Table of contents

PREFACE

2019 3rd International Conference on Artificial Intelligence, Automation and Control Technologies (AIACT 2019) was held in Xi'an, China from April 25 to 27, 2019. AIACT 2019 was co-organized by is organized by Xidian University and Hong Kong Society of Mechanical Engineers, sponsored by York University and Fudan University. The conference provides a useful and wide platform both for display the latest research and for exchange of research results and thoughts in Artificial Intelligence, Automation and Control Technologies and other topics. The participants of the conference were from almost every part of the world, with background of either academia or industry, even well-known enterprise. The success and prosperity of the conference is reflected high level of the papers received.

The proceedings are a compilation of the accepted papers and represent an interesting outcome of the conference. There were 332 submissions including 293 papers and 39 abstracts. After rigorous peer review, 116 papers and 15 abstracts were accepted. This book covers 3 chapters: Artificial Intelligence; Design and Applications of Artificial Intelligence; Automatic Control.

We would like to acknowledge all of those who supported AIACT 2019. Each individual and institutional help were very important for the success of this conference. Especially we would like to thank the organizing committee for their valuable advices in the organization and helpful peer review of the papers.

We sincerely hope that AIACT 2019 will be a forum for excellent discussions that will put forward new ideas and promote collaborative researches. We are sure that the proceedings will serve as an important research source of references and the knowledge, which will lead to not only scientific and engineering progress but also other new products and processes.

Prof. Dan Zhang, York University, Canada

Prof. Xuechao Duan, Xidian University, China

List of Chairman, Conference Chair, Program Committee, Advisory Committee and International Technical Committee are available in this pdf.

All papers published in this volume of Journal of Physics: Conference Series have been peer reviewed through processes administered by the proceedings Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.

It is a crucial problem to process abbreviation in the field of natural language processing. The most commonly used way to cope with this problem is to construct the reference database by predicting the abbreviation through its fully expanded form. Previous work on abbreviation prediction mostly rely on traditional machine learning algorithms, which inevitably requires a large number of manual annotations or expert knowledge to establish a feature system. In this paper, a neural network model based on CNN-BLSTM-CRF is proposed, which can predict Chinese abbreviations better without relying too much on the feature system: Firstly, convolutional neural network extracts phrase and Chinese character information from the fully expanded form, and then BLSTM-CRF deep network is constructed to annotate the fully expanded form, so as to extract its corresponding abbreviation form. The experimental results show that the method in this paper can perform better than the state-of-art method in traditional machine learning, and the results provide a reference for abbreviation research and the construction of resource repository.

An approach that employs deep learning technology is presented to recognize satellites based on radar high-resolution range profile (HRRP) data. We focus on extracting effective satellite recognition features in this paper. Thus, a deep learning model is constructed by gated recurrent unit (GRU) neural network and support vector machine (SVM) to extract more abstract and accurate features. Firstly, the radar HRRP data of four satellites is obtained by simulation. And data preprocessing has been done according to HRRP characteristic. Next, a GRU-SVM model is set up and some deep learning skills, such as dropout and cross validation, have been applied to improve recognition accuracy. The training results of GRU neural network show their effectiveness. In order to demonstrate the superiority of this approach, five other feature extraction methods have been used as a comparison based on clean satellite HRRP data and noisy data. The experiment results show that the presented GRU-SVM model could recognize satellites effectively and accurately, and has better recognition performance and noise robustness compared with five other methods.

Welding robots are widely used in manufacturing industries, and reasonable welding path planning is an important issue in production efficiency. Welding robot path planning aims to arrange the sequence of weld joints and find the optimal welding path for the robot, which is essentially a combinatorial optimization problem. With the development of artificial intelligence, swarm intelligence algorithms shed new light on this problem. Fireworks algorithm (FWA) is a newly proposed swarm intelligence algorithm, simulating the process of fireworks explosion producing sparks to find the optimal solution. It has shown excellent performance in continuous optimization problems. In this paper, a discrete fireworks algorithm (DFWA), is proposed to solve the welding robot path planning problem. We introduce some operations to the framework of traditional FWA. In DFWA, 2-opt local search and crossover operator are applied on the explosion sparks. A new way of fireworks quality judgment is designed. Mutation operator is implemented for the generation of Gaussian sparks, and the selection strategy is improved. Simulation experiments have been made to verify our method and compare its performance with other current swarm intelligence algorithms. Experimental results show that the proposed method performs well with good convergence, stability and accuracy. It is effective for welding robot path planning.

The recent technological advances in data gathering, embedded micro-devices and mobile networking has significantly advanced the applications of small-size and numerous tiny nodes. Large scale wireless sensor networks (LSWSNs) are intensively studied and used in the fields of traffic avoidance, intelligent family, medical diagnostic, environmental, multimedia surveillance, military affairs and so on. The recent success of emerging LSWSNs technology has encouraged researchers to develop new low energy clustering algorithm in this field. In LSWSNs, reducing communication energy consumption of sensor will not lead to maximize network lifetime for the total system. The low energy clustering is a typical NP-hard combinatorial optimization problem. In this paper, an immune adaptive cuckoo search algorithm (IACSA) is given to reduce total energy consumption. We first design a fitness function to evaluate energy consumption of system. The IACSA is designed to improve the energy efficiency for LSWSNs. It has the advantages of immune generator that takes into account different benefits and adaptive operator to enhance the convergence rate. Simulations are conducted to show a comparison of IACSA with the shuffled frog leaping algorithm (SFLA), particle swarm optimization (PSO) and artificial fish swarm algorithm (AFSA). Results show that the proposed IACSA has lower energy consumption compared to the SFLA, AFSA and PSO, which means that the proposed method reduces the energy consumption.

Recent advances in efficient software algorithms, electronics and networking technologies have promoted the development of low computational complexity, low-cost, low storage, and intelligent tiny nodes. Low power wireless sensor networks (LPWSNs) are composed of some sensing units with limited communications as well as sensing capabilities. LPWSNs are approaches in a good deal of domains such as traffic control, automated assistance for the elderly monitors and so on. Recently network lifetime optimization has been receiving a lot of attention for wide applications of LPWSNs. Achieving a longer network lifetime under the restricted power source has a large calculation difficulty, which can be regard as an NP-hard problem. An improved clone elite monkey algorithm (ICEMA) to determine the network lifetime optimization in LPWSNs is given. In this paper, in order to achieve maximum network lifetime with full coverage, we first build a system model to calculate a better network lifetime. It is designed to increment the lifetime of the nodes for LPWSNs. The ICEMA has many advantages by combining the clone strategy as well as elite strategy. The simulations verify the robust and efficiency of ICEMA when compared with strategies based on evolutionary algorithm (EA), particle swarm algorithm (PSO) and artificial fish swarm algorithm (AFSA) under a LPWSNs conditions. The outcomes demonstrate that the proposed ICEMA can achieve a longer network lifetime than EA, PSO and AFSA while taking the same computational complexity.

The rapid development of intelligent sensing, micro electro-mechanical systems and communication has made it feasible to equip low computational complexity, low energy consumption, autonomous, and intelligent sensor nodes. High-density wireless sensor networks (HDWSNs) have information acquisition and communication abilities. HDWSNs are widely used in a number of areas including traffic avoidance, homelands security, target monitoring and so on. One of the major challenges in HDWSNs is to maximize the point coverage percentage. It is important since it is known that obtaining an optimal coverage target for HDWSNs is an NP-hard problem. In this paper, we use an elite adaptive particle swarm optimization (EAPSO) to solve the issue of target coverage in HDWSNs. In order to improve the effectiveness of system, a system model is provided to evaluate the monitored rate for HDWSNs. The proposed EAPSO with an efficient particle swarm optimization in discrete mode has the advantages of both adaptive as well as elite strategy. Numerical simulations are conducted with a number of nodes and targets using EAPSO, evolutionary algorithm (EA) and simulated annealing (SA). In the simulations, a better performance of EAPSO is given when it is compared with EA and SA with the same computational complexity.

Class-incremental learning refers to the problem that the number of classes increases dynamically in the training stage. In this paper, a method based on one class support vector machine (OCSVM) is proposed for class incremental learning. The basic idea of the proposed method is that one class model is used to learn the distribution of new class, and then the knowledge learned and previous models are used for class incremental learning. The proposed method does not need to store all training data and makes full use of the learned information, so it has lower memory cost and time consumption. Also, the proposed method can build model when only one class is available. Compared with OC2MC, CIL, NN, the proposed method also achieves competitive results on different datasets.

Collaborative representation based classification (CRC) codes a test sample as a linear combination of all the training samples. However, the recognition rate of CRC is not ideal when available training samples are insufficient, as the test sample cannot be accurately represented by limited training samples. In order to address this issue, we propose a novel idea of training samples fragmentation. First, all training samples are divided into two new training sample sets according to the similarity among them. Next, the test sample simultaneously uses the two new training sample sets to perform CRC, which ultimately uses 2M "nearest neighbors" from the two training sample sets to represent and classify the test sample. In addition, this method also takes advantages of a new fusion classification mechanism based on histogram similarity and Euclidean distance, which has been proven to perform better than Euclidean distance classification. The experimental results reveal that the proposed method performs better in face recognition compared with the most representation based classification methods.

The increase of network users has led to a large number of commentary languages on various network platforms. Traditional manual processing is time-consuming and labor-intensive. We need a mechanized way to process these commentary corpora and quickly uncover the emotional tendencies. A method of sentimental orientation analysis of comment text based on deep learning is proposed. First, we used GloVe model to train the word vector. Then, Give the different weight on word vector by using TF-IDF. Finally, the processed word vectors would be classificated by TextCNN. Experiments were carried out on the six categories of commodity review data crawled by Jingdong. This method can effectively identify the emotional tendency of the review text, which is more accurate than the traditional deep learning method.

In order to solve the multi-objective problem, an improved quasi-oppositional multi-objective antlion optimization algorithm based on differential evolution (DEQOMALO) is proposed. This algorithm overcomes the defect that antlion algorithm is easy to fall into local optimum. On the one hand, this algorithm uses the idea of differential evolution to make full use of the information of the ant and the elite antlion to improve the position updating method of the original algorithm. On the other hand, the population is optimized by quasi-opposite learning strategy, and the original population and its quasi-opposite individuals are mixed and then selected as the new population, which greatly increases the diversity of the population. Finally, typical benchmarks are selected to compare the algorithm with the original antlion algorithm and other MALO algorithms with traditional evolution strategies. Experimental results show that both convergence and distribution of the improved algorithm are greatly improved. The proposed DEQOMALO algorithm has good adaptability and effectiveness in solving the two-objective optimization problem.

The quad-rotor UAV (Unmanned aerial vehicle) has a wide application market for its simple structure, easy operation and strong adaptability. During the flight, the endurance of UAV is an important parameter for flight planning, and it is of great significance to master the endurance capability of UAV. The endurance of UAV is mainly decided by the remaining capacity of the battery and the future energy consumption which changes with flight conditions. In this paper, a Flight Condition-Energy Consumption Model is established by analysing a large number of flight data with machine learning method, and the effects of different regression algorithms are compared. The validity of the model is verified by the actual flight. The endurance of the aircraft can be estimated using this model with the given remaining capacity of the battery and the future flight tasks.

Facial image based human age estimation is of great application significance. The public-available facial image datasets used for age estimation suffer greatly from the uneven distribution of images of different age groups, which may lead to the low estimation accuracy of the under-sampled age categories and limit the usage of the age estimation in certain applications. We propose a three-stage probability adjustment based CNN algorithm to solve the imbalanced distribution problem of the dataset. In particular, we construct an ENIN neural network structure by applying the Network in Network (NIN) structure to the traditional convolution neural network (CNN) and use the probability vector adjustment to improve the classification accuracy of the under-sampled age categories. Then, we filter out the images with high possibility of being misclassified after the probability vector adjustment and reset their categories by comparing cosine similarity and retraining the ensembled ENIN classifier. We also introduce a population-age-distribution based accuracy metric Accuracy-P to estimate the performance of the age estimation algorithm in real-world applications. Our experimental results confirm that our algorithm can effectively improve the overall estimation accuracy by significantly improving the accuracy of the under-sampled age groups while maintaining satisfactory accuracy for the other age groups.

The data generated from online communication acts as potential gold mines for discovering knowledge for end users. Large amount of data is also generated in the form of web documents, emails, blogs, and feedback, etc. Text analytics and opinion mining are used to extract human thoughts and perceptions from unstructured texts. This paper proposes a method that focuses on analysing different classification and clustering algorithms aimed at extracting and consolidating opinions of customers from social media sites like Facebook, Twitter and through surveys, at multiple levels of granularity to monitor and measure customer satisfaction. This is an automated approach, in which algorithms aid in the process of knowledge assimilation identification and the analytics. Domain experts ratify the knowledge base and provide training data sets for the system to intuitively gather more instances for ratification. The system identifies opinion expressions as phrases containing opinion words, opinionated features and also opinion modifiers. These expressions are categorized as positive, negative or neutral. Opinion expressions are identified and categorized using localized linguistic techniques. Opinions can be congregated at any desired level of specificity i.e. feature level or product level, user level or service level, etc. It has been found that J48 classification algorithm and simple k-means clustering algorithm are most suitable for restaurant customer reviews.

In the traditional medical system, individual medical data is managed by hospitals rather than individual patients. It is difficult to exchange effectively with fragmented storage, and large amounts of data are difficult to realize their potential value. With the rapid development of medical informatization, centralized storage of fragmented medical data has been unable to meet the relevant needs of the medical industry. To solve the difficulty of sharing and the complexity of confirming rights in the medical system, this paper proposes a medical data sharing model based on blockchain. The model provides reliable storage with IPFS file system, uses Proxy re-encryption to realize data sharing and ensure data proprietary rights, and uses Token economic system to measure the contribution in the sharing process, which stimulates the enthusiasm of sharing. At last, based on the existing sharing problem of medical data, the paper shows the potential solution.

Inconsistency of workpieces always results in low level automation of subsequent finishing process. As decreasing number of skilled workers, the demands of flexible manufacturing with robots is growing. In this paper, a kind of workpiece with complex free-form surface will be chamfered by industrial robots. Due to the intolerant differences among workpieces, traditional teach-playback method for robot is not appropriate for this application. The edge spline of each workpiece needs to be detected so that trajectories are generated to chamfer. Thus, an approach to edge extraction based on a 3D point cloud obtained by the 3D industrial camera is introduced to solve this problem. A time optimizing method is proposed to accelerate the extraction process. Finally, we decrease computing time from 55 minutes to 300 seconds, and the precision is 0.4180 mm.

The endpoint stiffness of the human arm has been long recognized as a key factor in the smooth contact between humans and environment. And the endpoint stiffness of the human arm is highly correlated with the surface electromyography (sEMG) produced by the contraction of the muscles. In this paper, the Gene Expression Programming (GEP) Algorithm is proposed to estimate the endpoint stiffness of human arm based on sEMG. This paper improves the traditional decoding method of GEP. Instead of generating an expression tree, it is directly decoded by looking for the effective length of the gene. And experimental results show that nonlinear models such as GEP models in this paper have higher correlation and lower RMSE (root mean square error) than regression stiffness using linear regression models. Selecting different feature of EMG signals, the correlation coefficient and the root mean square error of the model is very different. For the GEP model in this paper, WPTSVD (Wavelet Package Transform Singular Value Decomposion) and WTSVD (Wavelet Transform Singular Value Decomposion) are selected as the feature of sEMG signals have high performances and the correlation can reach 57%±12.1%

This paper integrates the current Google's most powerful NLP transfer learning model BERT with the traditional state-of-the-art BiLSTM-CRF model to solve the problem of named entity recognition. A bi-directional LSTM model can consider an effectively infinite amount of context on both sides of a word and eliminates the problem of limited context that applies to any feed-forward models. Google's model applied a feedforward neural network, causing its performance to weaken. We seek to solve these issues by proposing a more powerful neural network model named BT-BiLSTM. The new neural network model has obtained F1 scores on three Chinese datasets exceeds the previous BiLSTM-CRF model, especially on the value of recall. It shows the great value of the combination of large scale none-labelled data pre-trained language model with named entity recognition, which inspire new ideas on other future work.

This paper proposes an algorithm for perception English essay, in order to improve the ability of Chinese students' writing which expressing more standard. Because students are influenced by their mother tongue in second language acquisition, English expression is not accurate enough. From the lexical and syntactic aspects, this paper uses the language sense relevance analysis algorithm (N-LanSen) based on the N-model to deeply perceive the students' inappropriate expression in the essay. Different extraction methods are used to extract different features, and the corresponding feedback can be given to the places where the expression is not standard. The experimental results show that the algorithm can detect the place where English expression is not standard with high accuracy.

In view of the imbalance and even the abnormal phenomenon of reactive circulation of parallel high voltage transmission lines with similar measured parameters in actual operation, combined with field investigation and test, theoretical analysis and simulation demonstration, it is clear that this phenomenon is caused by the use of unconventional phase sequence and unreasonable transposition of the two-circuit lines when erecting the same tower. Based on the PSCAD simulation, the corresponding improvement measures are put forward from two aspects of optimizing the arrangement of phases and transposition. The results show that in order to solve the power imbalance problem of parallel coupled transmission lines, it is recommended to optimize the phase sequence of conductors in consideration of economy and operability.

For the nonlinear engagement between two pursuers and a single evader, in which small angle deviation assumption or other linearization conditions may not be applicable, a cooperative guidance law is analyzed and derived in the framework of a zero-sum two-person differential game. State-dependent Riccati-equation method is used to obtain the suboptimal solutions of the pursuit-evasion problem by choosing the line of sight angular rates as state variables, and closed-loop form guidance strategies that can be used online for the players are derived. Time-to-go estimations are not had to be taken into account to apply the guidance law proposed. Nonlinear two-dimensional simulations are carried out to validate the performance of the cooperative guidance law, and the comparison with a current cooperative linear quadratic differential game guidance law is made.

Based on the NaSch model, aiming at the characteristics of high-speed train tracking operation under moving block, a cellular automaton dynamics model is proposed to simulate the energy consumption of multi-train tracking operation on Railway lines. Through numerical simulation, this paper studies the influence of different line density and station stopping time on the energy consumption of high-speed railway traffic flow. The simulation results show that the energy consumption model can accurately reflect the energy consumption of multi-train tracking operation. At the same time, the phenomenon of traffic waves that sometimes go and sometimes stop is reproduced. According to the simulation results, it can be concluded that the energy consumption of multi-train tracking flow decreases gradually with the increase of line initialization density, and the average energy consumption of tracking traffic flow decreases with the increase of station stopping time. The results provide a scientific theoretical basis for efficient operation and energy-saving operation of high-speed railway. It has some guiding significance.

Aiming at the problems of insufficient network fusion and low detection efficiency in current object recognition using RGB-D images, a recognition algorithm based on the medium-level layer-by-layer fusion of dual-channel networks is proposed. First of all, the RGB and Depth networks are trained with ten labelled RGB-D indoor objects respectively, and then determine the fusion coefficients according to the identify accuracy of two types networks. Finally, two kinds of features are merged in convolutional layers step by step to obtain the final weights. By testing on the challenging NYU Depth v2 dataset, we found that the recognition accuracy of our method is 92.85%, and average detection time is 61.03ms per image. Through comparison experiments, we got the conclusion that average accuracy of the RGB-D layer-by-layer fusion network is 5.22% higher than that of the RGB network.

Spoken language understanding(SLU) is an important function module of the dialogue system. Slot filling and intent detection are two key sub-tasks of task-oriented spoken language understanding. In recent years, the methods of joint recognition have become the mainstream methods of spoken language understanding to solve slot filling and intent detection. Since deep neural network has advantages such as strong generalization and autonomous learning characteristics compared with traditional methods. So far, slot filling and intent detection have been developed from traditional methods to deep neural network methods, and the performance has also been significantly improved. This paper introduces the methods of two tasks from the independent model to the joint model. It focuses on the joint modeling methods based on deep neural network, analyzes current problems and future development trend of two sub-tasks.

The Flight Data Recording System (FDRS) records a lot of parameters of the aircraft during flight, which can be used for the test-flying, training mission of aircraft and so on. Effected by the working environment, information interference and its non-stability, the outliers and noise often exists in the FDRS data. These noises and outliers have a great impact on the use of FDRS. The aim of this paper is to remove outliers and de-noising of navigation data in FDRS. The causes of outliers and noise in FDRS data are analyzed firstly, with a reference suggestion proposed. Then the Letts criterion is used to remove outliers and the Modified Ensemble Empirical Mode Decomposition (MEEMD) is applied to achieve denoising for FDRS. Results demonstrate that outliers are removed and the navigation data are de-noised effectively.

Scholars pay more and more attention to the spoken dialogue system after the emergence of deep learning technology. The task-based dialogue system has become one of the most important branches in the field of spoken dialogue systems. Dialogue management is the core of the task-based dialogue system, and its research theory and technology have been gained extensive attention. This paper summarizes the research progress and current situation of task-based dialogue system and dialogue management strategy. Firstly, it summarizes the status of the task-based dialogue system models and compares their advantages and disadvantages. Then it focuses on the analysis of various dialogue management research strategies from the perspective of model theory and research methods. Finally, it looks forward to the future research direction of task-oriented dialogue system combined with dialogue management.

In order to improve the operational capability of heterogeneous combat networks (HCN), a decision center (DC) selecting method is proposed in this paper. According to the characteristics of modern information warfare, the entities in HCN are divided into influencer entities, decider entities, sensor entities and DC entity. The optimal distributed set of information flow links is obtained based on genetic algorithm. In light of the degree of each entity in this set, the entity with highest degree is selected as DC. Since DC has higher decision capability, highest degree means that the operational capability of majority operational chain is enhanced. And thus, the overall HCN operational capability is improved.

Clustering is a process of classifying data into different classes and has become an important tool in data mining. Among many clustering algorithms, the K-means clustering algorithm is widely used because of its simplicity and high efficiency. However, the traditional K-means algorithm can only find spherical clusters, and is also susceptible to noise points and isolated points, which makes the clustering results affected. To solve these problems, this paper proposes an improved K-means algorithm based on kurtosis test. The improved algorithm can improve the adaptability of clustering algorithm to complex shape datasets while reducing the impact of outlier data on clustering results, so that the algorithm results can be more accurate. The method used in our study is known as kurtosis test and Monte Carlo method. We validate our theoretical results in experiments on a variety of datasets. The experimental results show that the proposed algorithm has larger external indicators of clustering performance metrics, which means that the accuracy of clustering results is significantly improved.

Milk is one of the most common drinks in people's lives. Its quality is related to people's health and safety. Therefore, food regulatory authorities and milk manufacturers are increasingly strict with milk. The production date on the packaging is an important part of milk testing. One. Aiming at the problems of long time and high cost of manual detection, this paper proposes a method based on mathematical morphology for milk production date location and tilt correction based on computer vision related technology. The method firstly preprocesses the collected milk image, then uses the mathematical morphology method to obtain the candidate region of the milk production date, then analyzes the candidate region, extracts the accurate milk production date region, and finally uses the least squares principle. The straight line is fitted to calculate the tilt angle of the milk production date, and the milk image is tilt corrected using the rotation formula. In order to verify the comparison with the Hough transform method, the experiment proves that the proposed method can accurately predict the milk production date and tilt correction.

In the field of oil pipeline monitoring applications, the guarantee of high accuracy target detection is required. Aiming at the problem of moving target detection under static camera, a method of mean background model based on probability and statistics is proposed. The background model is quickly constructed. The background threshold is adaptively used to eliminate the influence of illumination changes. The target is detected quickly and accurately. Finally, the proposed target is proposed. The background update method updates the background model in real time. The simulation results show that the better detection results can be obtained under the influence of light changes. The background model has strong stability and adaptability, and can be applied to the real-time monitoring system of the field oil pipeline.

This paper studies the multi-agent systems consensus problem considering Lur'e-type nonlinearity under directed network. Compared with the conclusions of the existing undirected communication topology, the conclusions obtained have greatly reduced the communication topology requirements and are more general. This paper designs a control protocol based on the observer, which solves the problem that the system state information is unknown. By decomposition of a specific form of the Laplacin matrix, consensus problem is converted into stability problem for low-dimensional systems. In this paper, the method of solving the control protocol is given and proved by the piecewise Lyapunov function. The simulation results show that the designed control protocol can solve the consistency problem.

In recent years, the causes, prevention and control methods and prediction system of continuous haze weather have become the focus of social research.Taking the meteorological observation data of Hebei province as an example, this paper studies the influencing factors of continuous haze weather.Firstly, the improved TOPSIS algorithm is used to preprocess the data. Then, it combines the data mining methods such as hierarchical analysis and grey correlation analysis to carry out modeling analysis, and it is concluded that the particulate matter in the pollution source has the greatest impact on the haze weather. Moreover, as the aerodynamics factor affects the diffusion and aggregation of the haze pollution source, it is found that it has a linear influence on the formation of haze weather.Finally, through the current meteorological observation data, BP neural network is used to predict haze weather changes.A large number of experimental results show that, on the premise of allowing a certain error rate, the prediction effect of the BP neural network model is relatively accurate, and it also indicates that the formation of haze weather is closely related to the air quality index factor and meteorological index factor.

Aiming at the requirement of video surveillance and disease identification in modern greenhouse, this paper designed a system consisted by greenhouse intelligent monitoring and disease identification system. At first, this paper using OpenMV camera designed a disease monitoring system which could recognize and collect disease images automatically by multi-color thresholds tracking. Then, this paper designed a cuckoo search and BP neural network collaborative search (CSBP-CS) algorithm, the algorithm combined the global search capability of Cuckoo Search(CS) and back-propagation algorithm of BP algorithm to optimize weights and thresholds collaboratively. this paper took three tomato diseases and normal leaves as research objects, firstly step was to separate the disease spots from disease images, then was to extract 56 classification features and select 47 excellent classification features by relief F to construct CSBP-CS RF classifier. Finally, this paper compared the classification accuracy of CSBP-CS with CSBP-CS RF network and analyzed the effectiveness of relief F. The simulation results showed that the average correct recognition rate of the CSBP-CS RF was approximately equal to CSBP-CS under the same conditions, but CSBP-CS RF is more simple, so Relief F can help to speed up the efficiency of CSBP-CS.

In recent years, the neural network method can automatically learn effectively features. Unlike traditional discrete features, neural network features are mostly continuous features and can be automatically combined to build higher-level features. The efficiency of the features has been proven in numerous tasks in natural language processing and has led to breakthroughs. In this paper, we propose a event extraction system based on combination of multiple embedded features. Our work is mainly based on the three aspects: (1) traditional pipeline systems have serious error propagation problems; (2) there are several different event descriptions in the text; (3) representation learning can provide rich semantic and syntactic representation. As a result, we achieve competitive performance, specifically, F1-measure of 60.25 in event extraction. Meanwhile, evaluation results point out some shortcomings that need to be addressed in future work.

As the uncertainty output characteristics of wind power, the large-scale wind power integrated to the grid, has brought new challenges to power system operation. Based on the uncertainty characteristics of wind power, this paper establishes the wind power output model, then focuses on the positive effects of price-based demand response (PDR) and incentive-based demand response (IDR), which are allowed to participate in the power balance, on the unit commitment (UC) with high wind power penetration, and establishes the PDR and IDR models. Based on the chance constrained programming, this paper proposes a stochastic UC model with large-scale wind power integration considering demand side resources(DSR). Stochastic simulation combined with particle swarm optimization (PSO) algorithm was presented to solve this model. At last, the model is proved feasibly and effectively by testing the IEEE 10 machines system.

In order to improve the deployment efficiency of the microwave relay network, a method for determining access node of microwave relay network based on genetic algorithm is proposed in this paper. System analysis of network structure characteristics, node attributes and deployment process, the access node deployment issues is transformed into continuous area location issues. After discretization the area, the set of candidate access nodes is obtained considering boundary points and boundary chords. Comparing the coverage set of each access node, the candidate set is de-redundancy. Then taking coverage ratio maximum and the deployment cost minimum as the optimization goal, the optimal set of access nodes is obtained from candidate set by using genetic algorithm. A case with 50 user terminals is simulated by our method. Results show that after de-redundancy and genetic algorithm calculation, the number of access nodes is optimized from 240 to 17, and only 9 user terminals are repeated covered by different access node.

The application of industrial robots has greatly promoted the development of industry in the past decades. Now with the proposal and prevalence of Industry 4.0, industrial robots are required to be more independent and intelligent to accomplish more complex and flexible tasks. The advancement of industrial robots relies on the development and progress of multiple technologies, among which sensors are the indispensable part. They can acquire abundant information to help industrial robots implement their functions. This paper reviews the recent literatures and gives a summary and introduction of the commonly used sensors in industrial robots. Additionally, the applications of these sensors in diverse functions of industrial robots are also presented. Finally, the developing direction and challenges of industrial robots in the future are discussed in the last part of this article.

Negatively correlated search (NCS) is a latest intelligence algorithm which maintains multiple individual search processes in parallel and models the behavior of individual search processes using probability distributions. In this paper, an improved NCS inspired by particle swarm optimization (PSO) is proposed, namely PSO-NCS. This algorithm introduces the global best solution (gbest) and personal best solution (pbest) from PSO to calculate the Bhattacharyya Distance (BD) of individuals in the population, which greatly reduces the computational time. Experimental results on six 100 dimensional functions and 10 CEC2005 benchmark functions show that the proposed PSO-NCS has outstanding capability in solving the optimization problems and outperforms the conventional NCS and other methods on most of the test functions. The computational time has been greatly shortened, which saves about 65% in average compared to NCS.

Surface evenness affects vehicle safety, service life, and passenger comfort. The road surface with poor flatness will produce resistance to the vehicle driving, making the vehicle vibrate and bump, hindering the safety of the vehicle driving, and will also cause local water accumulation, speed up the road damage and so on. In order to make the vehicle safe, passengers feel comfortable, road life growth, road flatness detection is particularly important.In order to make the vehicle safe to travel, passengers feel comfortable, the road service life growth, and the detection of road surface evenness is particularly important. In order to overcome the shortcomings of the existing algorithms for pavement evenness, this paper has put forward a new fast algorithm that can be applied to the continuous vialog, and according to the elevation value obtains the evenness standard deviation σ. The results show that the algorithm is high reliability, which provides a reference for the upgrade of instruments and equipment, and makes up for the shortcomings that the continuous vialog cannot be tested quickly.

For the problem of inaccuracy and cumulative error of visual odometer, The research and optimization of real-time Simultaneous Localization and Mapping of indoor robot based on binocular vision are studied. Based on ORB-SLAM2, key-frame map is created. First, the ORB feature is extracted from each frame of the input image and matched by fast approximation nearest neighbour(FLANN). Then, perform the preliminary pose estimation using EPnP, and optimize it with bundle adjustment and key-frame maps. When the tracking fails, apply key-frame maps and bag of words model to relocate. Finally, for the input binocular image, the SGBM is used to solve the parallax and then the depth, which will be converted to radar format data to create a map. In the research and optimization of real-time Simultaneous Localization and Mapping of indoor robot based on binocular vision, propose a method of assisted positioning with key frame map, and a method of feature matching optimization and relocation, which combines various pose optimization to achieve the accuracy of the robot indoors positioning and map construction.

Practically grasp the global development trend of the basic research on intelligent robots and provide support for the scientific development of intelligent robots in China. Relying on bibliometrics, this paper has proposed a four-dimensional competitive situation analysis system for basic research based on activeness and talent distribution, etc., which can effectively analyze the global situation of basic research on intelligent robots, able to reveal and describe the competition status quo and situation of the basic scientific research in specific fields. There is a big gap between China and the United States in basic research on intelligent robots, mainly including such aspects as follows: uneven distribution of high-level basic research talents; insufficient international cooperation in high-level scientific research; lower research activeness; layout of key research and development.

Aiming at the problems of large work intensity, low intelligence level and low efficiency, the production date intelligent recognition algorithm was designed. The algorithm optimizes the image preprocessing algorithm, combines the grid statistical method and the projection density method to extract the feature vector, and uses the support vector machine algorithm to identify the production date. The experimental results show that the algorithm can accurately identify the production date and achieve the expected recognition effect. Has practical significance.

In this paper, we focus on the behavior of virus spreading on complex networks. A SIQR epidemic model on small-world network is established. We find the spread of infectious diseases threshold, which can determine the dynamic behavior of the system in a positive invariant set. If R₀>1, there are two equilibriums in the system, the disease-free equilibrium and the endemic equilibrium. If R₀<1, There is only the disease-free equilibrium in the system. It is globally stable in the positive invariant set. Furthermore, we analyze the equilibrium and its stability. Finally, some numerical simulation studies are provided to illustrate the effectiveness of the results.

Ensuring the normal operation of the transmission lines, which provides a path for directing the transmission of energy from one place to another, is a prerequisite for delivering power to cities and enterprises. A major threat comes from foreign objects, which may cause interruption of power transmission. Compared with traditional manual method, which not only consumes a lot of manpower, but more importantly, affects the safety and efficiency of power network, in this paper, we apply a neural detection of foreign objects for transmission lines. Transfer learning and data augmentation are used to mitigate data shortages. Experimental results show that even with small training data, the neural detection with transfer learning and data augmentation is an effective method for this task without loss of real-time property.

An accidental fall could do a great damage to the health of elderly. Failure to provide timely assistance after a fall may cause injury or even death. In this paper, a fall detection algorithm based on Convolutional Neural Network (CNN)-Long Short Term Memory (LSTM) combined network is proposed, which makes full use of the powerful feature extraction ability of CNN and the excellent time series processing ability of LSTM. Data required by the algorithm is only the resultant acceleration from a low cost three-axis acceleration sensor. The experimental results show that compared with the algorithms based on Support Vector Machine (SVM) and CNN, the proposed algorithm has higher detection accuracy with a small data volume, which is very suitable for Internet of Things (IoT) enabled fall detection applications.

In order to achieve low-cost, high-efficiency customized footwear design and service to meet the individual needs of customers, quickly producing various types of shoe last based on different needs has become the key to customization. According to the moulding characteristics of the longitudinal profile of the shoe last, the key characteristic parameters and morphogenesis logic are found. Combined with the parameterized nonlinear thinking feature, the parameter logic is constructed. Rhino's parametric design plug-in Grasshopper is used to develop the shoe last design process. The automatic moulding process of the sectional view, combined with the parametric design of the shoe last and the cross-section of the shoe last, can form a quick custom shoe last method. A professional and rapid design method is formed to enhance the design and production efficiency of customized shoe lasts.

With the rapid development of machine vision technology, more and more attention has been paid to the visual-aided inertial navigation system. It is important that to extract and track the line features at the dynamic situation in the visual-aided inertial navigation system which is based on visual line feature information to compensate attitude errors. A novel line feature description is proposed that use the SURF points to mark the LSD lines. Then, through coarse matching and fine matching, the function of continuously tracking the one line features in different images was realized. These line feature description and tracking method are applied in the visual-aided inertial navigation system, and its effectiveness is verified by the vehicle experiment.

In this paper, we present a simple but effective framework K-AFPN that incorporates feature pyramid method for small-size pedestrian detection, fully utilizing the lower-layer detail features and higher-layer semantic features. The method not only enhances the robustness of the features, but also improves the discrimination of the feature maps, achieving competitive accuracy. In addition, atrous convolution is used to optimize the network for high-resolution feature maps, avoiding information loss caused by frequent down or up sampling. On top of the backbone network, K-means algorithm is used to obtain optimal initial anchor base sizes, which reduces computational costs and improves location accuracy. Hence, our method pays more concentration on pedestrians, especially those of relatively small size. Comprehensive experimental results on two classic pedestrian benchmarks illustrate the effectiveness of the proposed approach.

The paper depicts an improved method for calibration parameters, which is a 2D homography, using point-line correspondences. The correspondences are the edge lines of the board provided by camera and the intersection points obtained by lidar. Particularly, a data preprocessing method is put forward and exploited to dramatically reduce the lidar's measurement error. Afterwards, we use a triangular board to establish constraints. At least 8 correspondences are employed to compute the initial value, and the final result is optimized by L-M algorithm. Massive experimental results show that proposed method has about 1 pixel improvement, compared with the previous methods.

Recently the most essential research in machine learning is to discover correct topics from large scale of electronic documents, and many studies about text modeling have been made in natural language process with the environment of big data. This paper discusses a fundamental way for people to form words while talking and describing ideas, the document is considered as words convergence and a new method for text modeling based on the relational analysis about verbs, nouns, and modifiers is proposed. Additionally the paper introduces a new algorithm for estimating parameters in the generative probabilities of the model, which can alleviate the process about statistics-based text analysis from computational complexity. Experimental results show that the method improves the quality for learning a document.

Deep CNN based semantic segmentation has been developed for several years and many models are proposed. However, most of them are designed for natural scene images such as PASCAL VOC, and cannot perform very well on remote sensing images, in which objects are much smaller and more densely distributed than those in natural scene images. In this paper, we demonstrate the importance of high-resolution feature maps and the problem of large dilated convolutional kernels in semantic segmentation of remote sensing images. Furthermore, we propose a Stepwise-Refined Large-Kernel Deconvolutional Network with a focus on small and densely-distributed objects such as houses and buildings, or long and narrow ones such as roads and rivers. Experiments on a public available ISPRS Vaihingen Challenge Dataset and our self-compiled Fujian Dataset show that our model outperforms the state-of-the-art models in semantic segmentation of remote sensing images.

The lift used in automobile maintenance industry is usually installed in a place permanently. It occupies the place, and also has poor adaptability to vehicles with different size. Furthermore, there exists physical connection between lifting columns, which affects maintenance operation. To solve these problems, a mobile hydraulic lift control system based on wireless network has been designed. It uses the STM32F103 microcontroller and nRF905 wireless communication module to form a control network for several mobile lifts. The height of each lift is measured with laser displacement sensors, and synchronously controlled through the wireless control network. The control system was verified by lifting the Jinlong bus with wirelessly connected four lifts. The lifting process is stable, which demonstrates a bright prospect for practical usages.

This paper develops an end-to-end neural network model for text-to-speech (TTS) system based on phoneme sequence. Inspired by the Tacotron-2, the proposed model adopts an encoder-decoder model with attention mechanism and applies mel-spectrogram to measure the intermediate acoustic feature. Phoneme sequence is used to replace the character sequence in order to overcome the shortage of the character feature used in Tacotron-2. Unlike the conventional concatenate methodology based TTS system, our model can generate waveform directly from phoneme sequence. In addition, analogue from text analysis, a new analysis methodology is proposed for phoneme analysis. Experiment result on LJ Speech dataset shows that, compared with char-based model, our model can get a comparative or better performance.

Recently, Lane departure warning system has attracted great attention as it contributes to vehicle active safety. In this paper, a vision-based lane departure warning system under high-speed driving is proposed. The system consists of two functional parts: lane markings detection and vehicle departure identification. The Hough Transform is applied to detect lane boundaries, which is a most effective detection method with high reliability. Based on the road line, a method using several Euclidean-distance-related parameters to calculate vehicle's position and its deviating status is proposed, which addresses the problem of efficient detection of lane departure under high-speed driving condition. The algorithm is tested and verified in various real driving conditions and proved a reliable and steady performance.

Industrial process anomaly detection mechanisms have been proposed to protect industrial control system to minimize the risk of damage or loss of resources. In this paper, an one-class Support Vector Machine based extended boundary (EB-OSCVM) is used to detect anomalies in industrial multivariate time series data from a simulated Tennessee Eastman Process (TEP) with many cyber attacks. In detail, determine the change points of each process variable and capture the causality relationship between the variables based on the location and time delay of the change points. Then, by monitoring the leaf nodes in the causality graph, we can know whether the system is abnormal, it can effectively reduce the dimension of process data. The EB-OSCVM extend classification boundary of OCSVM in order to reduce the error of noise, if data is outside the boundary of EB-OCSVM, there is an anomaly. Finally, tracing the anomaly source according to causal direction. An experiment is used to verify the effectiveness of the proposed approach, the results demonstrate that the approach presents a high-accuracy solution and traces the source of anomaly correctly.

Object detection algorithms based on convolutional neural networks are generally suitable for static gesture recognition. For actual hand gesture scenes, dynamic gestures are also widely used. A dynamic hand gesture recognition algorithm based on Channel State Information (CSI) and You Only Look Once: Version 3 (YOLOv3) is proposed for continuous dynamic hand gesture recognition. The data acquisition adopts a CSI-based radio frequency method. The adaptive weighted fusion, Kalman filtering, threshold segmentation and data conversion are used to generate gray value images. Finally, the YOLOv3 object detection algorithm is used to train and identify the grayscale image which include the information of continuous dynamic hand gestures. The effectiveness of the proposed method is verified by the recognition confusion matrix. And the proposed method has an average recognition accuracy of 94% for four custom dynamic hand gestures.

Human interaction with mobile robot becomes a popular research area and its applications are widely used in industrial, commercial and military fields. A two-hand gesture recognition method with depth camera is presented for real-time controlling the mecanum wheeled mobile robot. Seven different gestures could be recognized from one hand for mobile robot navigation and three gestures could be recognized from the other hand for controlling the gripper installed on the robot. Under the proposed control scheme, the mobile robot system can be navigated and can be operated at the same time for achieving missions by two different groups of hand gestures. The accuracy of the gesture recognition is about 94%. During mobile robot control experiment, the system works timely, accurately and stably for certain tasks such as directional movement, grasping and cleaning obstacles.

Target tracking is often used in tasks such as video surveillance. Its purpose is to obtain the target of interest in the sequence image accurately, robustly and in real time, and to establish the connection between the moving objects in each image. In recent years, with the continuous development of the global shipping economy, ships have gradually developed into large-scale and high-speed, and how to ensure the safety of ships sailing in these sports has become an important issue. Based on the principle of particle filter and mean shift algorithm, this paper takes the moving ship as the research object and uses the polynomial fitting method to analyse the performance of the two algorithms. A large number of experimental results show that the mean shift algorithm has higher accuracy, and the particle filter method has better real-time performance. And as the number of particles increases, the accuracy of particle filtering will gradually increase.

This paper takes multi-dimensional force measurement of lateral force power unit as research objective, and a piezoelectric multi-dimensional force measuring device with piezoelectric quartz crystal as force sensitive component is designed. First, the mathematical model between sensors' output and measured lateral force is established by theoretical derivation. The relationship between the force applied to measured object and measuring device's output is obtained through the finite element analysis, which verifies the measurement performance of measuring device. On the other hand, through ANSYS model analysis, a conclusion that the first natural frequency of measuring device is 1467.8Hz can be obtained, and the dynamic performance of measuring device is good, which enables accurate measurement of real-time varying lateral force.

Spoken language understanding is an important part of the human-machine dialogue system, intent detection is a sub-task of spoken language understanding, and it is very important. The accuracy of intent detection is directly related to the performance of semantic slot filling, and it is helpful to the following research of the dialogue system. Considering the difficulty of intent detection in human-machine dialogue system, the traditional machine learning method cannot understand the deep semantic information of user's discourse. This paper mainly analyzes, compares and summarizes the deep learning methods applied in the research of intent detection in recent years, and further considers how to apply deep learning model to multi-intent detection task, so as to promote the research of multi-intent detection methods based on deep neural network.

In combination with the requirements of the Tracked Vehicle-Cable Salvage project, which is aiming at the characteristics of the submarine crawler slip, the impact of the slip rate on the tangential drive to steering radius and steering angular velocity of the submarine crawler chassis. it is analyzed from the theoretical and experimental levels. Simplify the force model of the working submersible during steady-state steering, and it can obtain the steady-state steering body dynamics equation; in order to solve the arbitrary shear rate and shear displacement of the grounding surface in the steering process. According to the steering geometry relationship, it can combined with the empirical experience of seafloor sediments. The model derives the longitudinal driving. lateral resistance and drag torque of the left and right side of the track when it is during steady-state steering; when we analyzes the instantaneous lateral shift of the track shear rate during steady-state steering; it derives the instantaneous rate of the instantaneous steering rate, which based on the geometric relationship of the steering motion The relationship between the offset and the steering radius of the body, the steering angular velocity, and the sliding rate as a function of the steering radius and steering angular velocity of the body; when the steering model test is carried out to verify the influence of the slip rate on the steering radius and the steering angular velocity; The slip rate has an effect on both the steering radius and the steering angular velocity. It was measured steering radius becomes larger, the steering angular velocity becomes smaller, and the theoretical prediction curve is basically consistent with the measured data.

Based on the demand for deep-sea submarine cable salvage in Northeast Asia, a mathematical model of Sub Marine Tracked Vehicle-Cable Salvage (SMT-CS) was designed in this study. Considering the slip characteristics of the submarine track, the mechanical model of the SMT-CS during steady-state steering was simplified to obtain the kinetic equation of the vehicle body during steady-state steering. According to the steering geometrical relationship, the arbitrary shear rate and shear displacement of the ground plane during the steering process were solved. Combined with the empirical shear model of bottom sediments, the longitudinal driving force, lateral resistance and resistance moment of the larboard and starboard were derived during steady-state steering. The functional relationship between the lateral shift of instantaneous center of the conventional steering rate and the turning radius and angular velocity of the vehicle body as well as that between the slipping rate and the turning radius and angular velocity of the vehicle body was derived according to the geometrical relationship of steering motion. These are of great reference value to the design, performance prediction and maritime application of SMT-CS in the future.

Aiming at the axial imbalance force when the traditional axial piston motor works, the force balanced axial piston motor utilized the concept of double stator to improve the structure of the traditional axial piston motor. Because of the special structure, the motor can not only achieve the axial force balance, but also output a variety of torque and speed. The working efficiency of the motor was an important indicator for evaluating the performance of the motor. Therefore, the power loss and leakage of the motor were analyzed in detail, and the prototype was tested. The experimental results showed that: (1) When the corresponding two single motors in the motor worked at the same time, theoretically, the leakage between the distribution cylinder and the cylinder accounts for the largest percentage of total leakage, reaching 56%. When the whole motor worked normally, the total leakage reaches 9.55L/min theoretically, and the leakage between the plunger and the plunger hole accounts for the largest percentage of the total leakage, which is 44% of the total leakage, and the distribution cylinder and cylinder the percentage of leakage between bodies has decreased, but its specific leakage has not changed. (2) The volumetric efficiency of the motor decreased as the pressure at the inlet of the motor increased, and the mechanical efficiency and overall efficiency increased accordingly. (3) When the motor inlet pressure was the same, the volumetric efficiency of two single motors at the same time was lower than the volumetric efficiency of the entire motor.

Event-triggered group consensus problem is addressed in this paper. Based on the assumption named the in-degree balance, a directed first-order multi-agent system is constructed, which consists of two subgroups. To analyze the problem, the Laplacian matrix of the initial system is decomposed to be two matrices with less row and column dimension. Then the initial state vector can be substituted by a new one, each element in which is the error of two adjacent elements of the initial state vector. Then a new reduced-order system is constructed and the consensus problem is converted to a stability problem. With the designed Lyapunov function, the validity of the proposed algorithm is proved. The lower-bound of the inter-event time interval is also presented. Simulation results are presented to demonstrate the effectiveness of the protocol.

The offline programming technology of industrial robots is more and more widely used because it can effectively improve work efficiency. A DXF drawing file is a file generated by CAD that can serve as a bridge for data exchange with other software. Through experiments, this paper combines CAD-generated DXF drawing files with ABB robotic arm to realize data extraction of CAD drawings, and automatic generation and storage of robotic arm running programs, which are imported into the robot arm, so that the robot arm can be made according to CAD. The trajectory on the drawing performs the welding function. The simulation of path planning was carried out under the Robot studio software developed by ABB Robotics to meet the experimental requirements. The realization of this off-line programming system achieves the following two purposes. One is to assist non-control professional teaching, to help understand the simulation and actual production and processing of industrial robot arm; the second is to improve the efficiency of industrial robot arm programming and realize full automation programming.

This paper presents the use of forced convection heat transfer approach to near-space thermal environment of hypersonic aerodynamic ground simulation test method for surface heat flux with a great gradient and temperature gradient of the tip wedge (rn=1.5mm) shape component level thermal test. In the test device test section, two different high subsonic speed airflow were nested within each of the two vents from the exhaust, the impact of the specimen heated to test sharp wedge surface heat flow and temperature distribution in line with a hypersonic flight state aerodynamic heat distribution patterns. Numerical results show that under this scenario, the specimen rear stagnation point heat flux and heat flux can be adjusted two ways to adjust the air flow; test methods within a certain accuracy-related parts for the superb aerodynamic vehicle thermal environment simulation.

In order to realize the co-construction, sharing and intensive management of millions of stations across the network, supporting 5G go into operation efficiently, China Tower Group has developed a new generation of intensive and distributed operation maintenance monitoring platform over the inherited mobile base stations from the three major telecom operators. Based on the development and the application of this platform, FEU (an intelligent Power and Environment Monitoring Unit) is proposed, which uses the dynamic shared library method to embed the edge gateway to realize the edge computing capability of China Tower's intelligent operation and maintenance platform. It effectively solves the normalized monitoring of heterogeneous FSU and massive sites, constructs the intensive operation maintenance capability for China Tower and lays a solid foundation for the tower-based IoT [1, 2] expansion business.

The network traffic prediction plays the key role in congestion control and bandwidth allocation. A variety of traditional learning models such as artificial neural networks (ANN) have been applied in prediction. To avoid the drawbacks of traditional models for prediction, a novel robust minimax probability machine (RMPM)-based traffic prediction method is proposed in this paper. The prediction performance is tested on two different types of traffic data, Ethernet data flow and MPEG4 video flow, at the timescale 1. The experiments demonstrate that the proposed method attains satisfactory performance in prediction accuracy. Therefore, the proposed method can be used for congestion control or bandwidth allocation, to meet the user QOS requirements.

With the continuous progress of bridge technology and the requirements for bridge aesthetics, cable technology is increasingly widely used in long-span bridges. Due to the long-term environment of alternating stress, corrosion and wind-induced vibration, it is easy to cause local fatigue and damage of the cables, which endangers the safety of the entire bridge structure. The sleeve cable force sensor based on magnetic flux method is widely used in practical engineering among many measurement methods. Existing sleeve cable force sensors need to be installed at the beginning of construction, so it is difficult to reinstall them in the follow-up, and its long-term performance is also difficult to evaluate and measure online. To solve this problem, this paper introduces the design of a kind of magnetic flux cable force sensor with by-pass excitation, and presents its measuring principle and design structure. The designed by-pass excitation cable force sensor is expected to solve the problem of the existing sleeve sensor.

This paper introduces a multi-link robot, named treatment table robot (TTR), which has been developed as an important part in protan therapy system SC200. TTR is aimed to carry and move patient's body during treatment, so as to change the space location and poseture of patient'tumour to ensure the proton beem can always focus on it. To realize the requirement, the TTR has been designed as a 6-DOF robot. Based on the builded D-H model, the forward and inverse kinematics are studied. Then the joint trajectory planning has been researched. Finally, ADAMS software was enployed to simulate the kinematic of the TTR, which verified the feasibility of the kinematic algorithm.

This paper introduces a kind of management and control system of machining production line for aero-engine nozzle. The main logistics is based on the logic control of PLC, which can dispatch a robot to realize the automatic transfer of the parts between workstations, automatic feeding and unloading of processing and testing equipment. Workstations are equipped with RFID to detect the identification of each nozzle. The system can reconstruct the process route according to the nozzle type, associate processing and test data of the nozzle with their identification number to establish the processing file. It realizes data exchanges between management and control systems by OPC interface, achieves centralized monitoring and control of the equipment of the whole line and improves the informatization level of workshop.

In this paper, based on the problem of poor operation of sweet potato harvester, the discrete element method and TRIZ theory are used to optimize the key mechanism. Based on the discrete element method, EDEM software was used to simulate the operation process of the potato soil conveying and separating mechanism, and the optimal combination of working parameters was obtained. The linear velocity of the rod lift was 2.3 m/s, the inclination angle was 18 degrees, and the amplitude was 8 mm. Field experiments were carried out to verify the results. Through the problems in simulation and field experiment, based on TRIZ theory, the improvement scheme is proposed by using physical contradiction, and the two-stage potato soil transport separation mechanism is obtained. The simulation results show that the operation effect is improved obviously. Finally, the optimization of potato soil transport and separation mechanism is completed by determining the optimal working parameters and improving the structure innovation.

The commonly used method for measuring the density of a selective laser sintered powder bed is to extract a certain volume of powder bed powder, measure the mass of the powder, and then calculate the density of the powder bed using the mass formula. Aiming at the shortcomings of the existing methods, such as complex operation and low measurement accuracy, this paper proposes a method based on plumb bob method for selective laser sintering powder bed density measurement. The method can measure the density of the powder bed without causing large damage to the powder bed, and the measurement efficiency and precision are higher than the existing measurement methods. In this test, the most commonly used ABS powder was used, and its density was calibrated using the plumb method. The function of the depth between the depth of the plumb bobbin and the density of the powder bed was fitted to the density of the powder bed. The measurements provide data support.

Aiming at the problem of poor gathering effect of seedling transplanting combination mechanism that is designed independently, the intensive study has been done by using the main tools of TRIZ: the functional analysis, the physical contradiction and technical contradiction, the Su-Field model. And four feasible solutions have been found during the intensive study. Comparing the advantages and disadvantages of solutions and Taking the actual engineering situation into consideration, the final solution is obtained by collecting the advantages of the former solutions. The final solution can solve the poor gathering problem very well.

In the pavement design, the tire load is usually assumed to be a circular or rectangular evenly distributed load, which is actually not like this. In order to provide data reference for the development of tire-pavement contact stress intelligent test instrument, it is necessary to model the tire-pavement system, and construct tire-pavement 3D finite element model using large finite element software. Firstly, establish three-dimensional model of tire and road surface, and impose certain loads and constraints. Secondly, the triaxial contact stress of the tire-pavement under static conditions is analyzed. Then the triaxial stress under traction conditions is analyzed, and the triaxial stress under different conditions of use of the tire is compared and analyzed. The results show that as the load of the tire gradually increases, the maximum vertical contact stress and the maximum longitudinal stress gradually decrease, but the maximum lateral pressure gradually increases. When the inflation pressure of the tire is gradually increased, the contact stresses in the three directions are correspondingly increased. However, the maximum vertical contact stress and maximum longitudinal stress increase more obviously than the maximum transverse contact stress. In addition, the transverse contact stress could be ignored under traction conditions.

In recent years, robot arms are being developed for self-reliance supporting for those who are unable to move without using an electric wheelchair or live a bedridden life. For example, mounting a robot arm on an electric wheelchair or bedside could help the limb disabled people to perform daily activities without nursing. However, the existing robot arms are mainly using traditional operation interfaces like keyboards or joysticks. Under different scenarios, some patients with upper limb disabilities such as contracture cannot operate such interfaces smoothly. In this study, in order to reduce the difficulty in operating robot arms, we developed a new kind of operation system using smartphone touchscreen as screen joystick operating interface. Meanwhile, the system includes a filter which detects and adjusts maloperation due to the capacitive touchscreen panel with high maloperation rate. Finally, an operating test was conducted by a patient with upper limb disabilities, and the effectiveness of this study was shown from the evaluation of the trajectory of robot hands from the initial position to the aim position. In the future, we are planning to develop an innovate operation system for a robot arm which is mounted on the wheelchair, with the effectiveness confirm by several operation tests.

Remote meter automatic reading and smart meter reading is a vital part of smart grid and has been widely concerned. Traditional manual method is manpower consuming. Developing an automatic electricity ether reading system is necessary in practice. Two particular aspects in this task should be concerned. One is different-perspective problem, which causes by the variations of camera position and leads to pool display localization. Another is information diversity problem. Namely, the electrical meter used in this paper needs to recognize effective information from diversity information. This paper describes a novel automatic electricity meter reading system, which can automatically locate and recognise effective information. For the camera perspective variability problem, we utilize a new localization method based on neural network. Via actual meter monitoring experiment, the result shows our system can efficient recognition and achieve 97% accuracy.

In this paper, a two-neuron system with inertia and delay is proposed firstly. a PD controller is then applied to the system for the purpose of improving its dynamical performance. Through the mathematical transformation, we extend the system to a four-dimensional one with only time delays. With the help of the associated characteristic equation of the mathematical model, suffcient conditions for ensuring the system stability are proposed. Furthermore, with the time delay as the bifurcation parameter, relevant requirements for the generation of Hopf bifurcation are derived. Then a series of numerical simulatiosns are carried out to justify the theoretical analysis and it is found that the application of PD control scheme helps to advance the bifurcation point dramatically through a slight adjustment of the controller parameters.

Coal gangue separation is of key important for green mining. Robotic separation of coal and gangue based on the machine vision system, which is called a coal gangue picking robot with a gangue grab driven by four cables, is developed in this study. Here, this paper focuses on the equivalent position workspace, within which a coal gangue picking robot possesses the identical stability, to find out any more information about the structural stability for a coal gangue picking robot. First, the kinematic and kinetostatic models of the coal gangue picking robot are presented for analyzing the effects of it on the structural stability for the coal gangue picking robot. And moreover, a non-iterative polynomial-based optimization algorithm with the proper optimal objective function is presented based on the convex optimization theory, in which the cable with the minimum cable tension at any pose is determined. Then, three position performance indices are proposed to show the important effects on the structural stability for the coal gangue picking robot in a specified region of the workspace. Besides, a new workspace, the Equivalent Position Workspace (EPW), is introduced. Finally, the theoretical relationship between the two performance indices and the stability is corrected based on simulation results. The research has important guiding significance and practical value for coal gangue robotic separation.

The future power system can be considered as an aggregation of controllable distributed systems devices that coordinate with each other through the Internet of Things (IoT), energy internet (EI) new paradigms arise. In EI environments, agreement among multiple computers is essential for power generation, transmission, distribution, and consumption etc. applications. Coordinated control and security protection play important roles in the application of a distributed computing environment. A consensus approach can provide fault tolerance and prevent system errors and attacks, thereby providing the system with strong security in distributed computing environment. In this paper, a distributed Kalman filter with consensus coordinated control algorithm for multiagent systems (MAS) was proposed to investigate the energy internet with links failure by cyber attack. The proposed algorithm is evaluated in the energy internet through a MAS model using MATLAB software.

This paper puts forward a novel tandem ducted fan flying robot which has nice passivity and compact structure aiming at operating in complicated and dangerous environment. The structure of its ducted coaxial twin-rotor with reverse rotation is unconventional and make its aerodynamic characteristics much more complex. To solve the problem, the modelling method of the traditional open rotors with the blade element theory and the momentum theory is extended to the ducted coaxial twin-rotor structure, and the integrated dynamic model of the ducted fan flying robot is established. After the modelling, a static lift experiment has been carried out, and the aerodynamic model of the ducted coaxial twin-rotor structure is modified based on the test results.

For the interception problem of maneuvering targets, considering impact angle constraint and first-order autopilot lag, an adaptive integral sliding mode guidance law is designed. A new nonsingular finite-time integral sliding mode surface is constructed, which consists of LOS angle, LOS angular rate and LOS angular acceleration, and ensures that the states of the guidance system converge to zero on the sliding mode surface in a finite time. Moreover, an adaptive law is designed to estimate the upper bound of the target's unknown acceleration. The Lyapunov stability theory proves that the guidance system can converge to zero strictly in a finite time with the proposed guidance law. Finally, the simulation results verify the effectiveness of the designed guidance law.

In this paper, in order to comprehensively consider the economy and comfort of the special duel-axis-parallel PHEV, a vehicle control strategy composed of a shift strategy based on dynamic programming(DP), a rule-based energy management strategy and a dynamic coordinated control strategy based on motor torque compensation is proposed. A vehicle model which can show the dynamic characteristics of shifting and engine start – stop process is constructed for simulations. Shift strategy is extracted from the optimization results obtained by dynamic programming(DP). The energy management strategy takes the fuel economy of engine as a major consideration and the rule is determined by characteristics of engine, motor and battery. In the dynamic coordinated control strategy, an open-loop motor torque control based on clutch torque estimation is presented. Simulation is carried out on typical driving cycles. And the results indicate that the proposed vehicle control strategy can improve the fuel economy and reduce longitudinal impact caused by shifting and operated mode changing.

With the rapid development of solar photovoltaic generation, the effective prediction of photovoltaic is of great significance to mitigate its impact on power system. According to the analysis of main factors which affect power output of photovoltaic system, a short-term power forecasting model based on back propagation(BP) neutral network and LVQ-PSO-BP neural network and Markov chain method was established. The weather is clustered and distinguished by using learning vector quantization(LVQ) and the particle swarm optimization(PSO) is used to optimize BP neural network weights and thresholds, improving forecasting network training speed. Finally, daily predictive value is corrected by Markov chain method to improve short-term photovoltaic generation forecasting precision. The simulation results indicate that the proposed method can accelerate the speed of searching optimums, improving the classification accuracy of weather types and the precision of the photovoltaic generation output effectively.

The optimal control problems of exponentially damped system are considered in this paper. Exponentially damped system involves convolution integrals over exponentially decaying functions, which is used as damping models in viscoelastically damped structures. The dynamic constraint is in the form of a differential equation that includes integer derivatives and the "integral" term. The performance index considered is a function of both the state and the control variables. The traditional state-space approach is extended to the optimal control problems of exponentially damped system, A direct numerical technique is used to solve the resulting equations. Numerical simulations are provided to illustrate the above control design.

The rendezvous of UAVs as a first step in the formation of a UAV is crucial. In this paper, a maneuvering system based on virtual leader is proposed. Firstly, the relative kinematics model of the virtual leader and UAV is established, and then a proportional guidance law with falling angle constraint is designed to make the UAV. Finally, the desired attitude angle can be achieved, and the speed of the UAV can be controlled, so that the UAV can track the virtual leader at a set speed and maintain a relative distance, thereby achieving the purpose of formation flying. The simulation results show that this method can effectively guide the formation of UAVs and has certain engineering application value.

The two-axis pan-tilt platform (PTP) has been widely used in industrial control systems. The control performance of the PTP is sharply affected by the load variations, which are more significant on the pitch axis. Moreover, the PTP control system is sensitive to sensor noise and the friction moment. To achieve rapid and precise control, we build a pan-tilt platform model using the method of multiple system identification and design a speed controller and position controller on both the yaw and the pitch axes. The proposed method avoids the need to measure a large number of mechanical and electrical parameters. The experimental result shows that the proposed control system has small overshoot, fast response and suitable robustness. Relative to a controller designed and adjusted by experience, the designed system increases the bandwidth by approximately 50% on the yaw axis, and the settling time is decreased.

The single-hidden-layer neural networks (NN) has been widely used for complex system identification. However, the hidden neurons are often determined by trial-and-error method and the amount of neurons is usually large. This commonly leads to over-fitting problem and the training process is time consuming. In this paper, we propose a two-stage backward elimination (TSBE) method to obtain a parsimonious network with fewer hidden neurons but remains a good performance and saves training time. In the first stage, neural networks with a predetermined number of hidden neurons is trained based on stochastic gradient decent (SGD) algorithm with part of training data and Least absolute shrinkage and selection operator (Lasso) is applied for dropping redundant neurons leading to a simplified neural model. In the second stage, the remaining training data is used to update the parameters of the simplified neural model. A simulation example is used to validate and show that the novel approach gives a more compressed model and higher level of accuracy comparing with the recently proposed pruning-based method.

In this work, a disturbance observer is designed to obtain the total disturbance including the unknown part, and the error system of the disturbance observer is proved to be finite-time convergence. Based on the total disturbance, a sliding mode controller with exponential reaching rule is proposed for the speed control system of a marine diesel engine. The performance of the proposed controller and the disturbance observer is tested by the simulations on a marine diesel engine model. It shows that the proposed disturbance observer can achieve satisfactory results, and both the control performance and the robustness of the DOSMC are superior to the PID controller.

This paper provides and validates a class of weighted H_∞ output feedback controllers of switched LPV (Linear Parameter Varying) systems based on the MDADT (Modal Dependent Average Dwell Time) method. First of all, the article uses a multi-lyapunov function to convert the linear parameter varying system into a polytopic model, and designs a switching law that can guarantee the system to be globally uniform and exponentially stable on the MDADT method. It can be shown that when the state is stable and can be measured in real time, the switching law and weighted H_∞ controllers for each subsystem under the MDADT limit can be calculated. Finally, a numerical example is given to verify the feasibility of this method and the stability of the switched LPV system.

Recently many complicated reading comprehension models have been proposed. Most of these models are constantly improved from a basic model. However, most of the innovations are at the interaction layer. In this work, we analyse the shortcomings of Dynamic Memory Networks and its extended model. Then we decide to put forward some different opinions that making some improvements in the embedding layer based upon these models. First of all, this method uses pre-training to get character embedding. Our model uses CNN to train the character vector, and then we splicing the character vector and the word vector. Meanwhile, we validate the model with the bAbI dataset. Our results show that this method can improve the interaction between context and question words.

This paper proposed an original fractional-order proportional-derivative (PD) feedback controller which is designed to control the Hopf bifurcation caused by the congestion control system. The proposed $P{D}^{\frac{1}{n}}$ controller has the different order with the original congestion system. The proposed fractional-order PD controller has the different order with the controlled system and the communication delay is selected as the bifurcation parameter. Then the conditions of the stability and Hopf bifurcation are obtained by analyzing its characteristic equation and the stability domain can be extended under the adjustment of appropriate control gain parameters and the order. Therefore, the congestion system becomes controllable and the desirable behaviors can be realized. Finally, numerical simulations are carried out to testify the validity of the theoretical analysis in the designed fractional-order PD controller.

Aiming at the fully coupled problem of steady state and dynamic performance of a PI controller with one degree of freedom, two kinds of two DOF PI controllers were designed based on the setpoint filter structure and the setpoint feed-forward structure, and the design method of compensation link was introduced.On this basis, a variable structure two-degree-of-freedom PI controller was designed with reference to the idea of variable gain PI controller.Finally, the simulation results show that the controller can not only simplify the process of setting control parameters, but also take into account the steady-state performance, dynamic performance and anti-jamming performance of the system, and realize the function of output fast, no overshoot, no error tracking input signal changes.

Aiming at the poor performance of IP control system in tracking continuously varying input signals, a variable structure PI controller was proposed, which can eliminate overshoot of step response and track continuously varying input signals without error. The controller performs as an IP controller for step input signals and a PI controller for continuously changing input signals. Firstly, the control performance defects of traditional IP controller were analyzed, and then the structure of variable structure PI controller was built. Finally, the simulation results show that the VSPI and IP control systems have the same control effect when the step signal is input. Compared with PI control system, the VSPI and IP control systems not only have no overshoot but also have better response stability. When the input is a continuous sinusoidal signal, VSPI and PI can track the input signal almost without error, and the control effect is obviously better than IP control. The anti-jamming performance of the system is inversely proportional to the bandwidth value, but has nothing to do with the control mode.

In view of the uncertainty of data transmission due to the irregular data flow in the network and the limitation of network bandwidth resources, which makes the control performance and stability of FlexRay network decrease when data is transmitted at high speed and the reliability and safety of the control system can not be guaranteed, this paper proposes a method based on the Levenberg-Marquardt (LM) algorithm for neural network predictive control of the FlexRay bus. The method can predict the next moment operating state of the car network according to the current moment working state of the FlexRay car network to adaptively adjust task workload to adapt to changes in vehicle network system load and improve the reliability and stability of FlexRay network control system. The simulation results show that the neural network predictive control has good adaptability and robustness, which improves the control performance of the FlexRay automotive networked control system effectively.

Visual Simultaneous Location and Mapping (SLAM) based on RGB-D has developed as a fundamental capability for intelligent mobile robot. However, most of existing SLAM algorithms assume that the environment is static and not suitable for dynamic environments. This is because moving objects in dynamic environments can interfere with camera pose tracking, cause undesired objects to be integrated into the map. In this paper, we modify the existing framework for RGB-D SLAM in dynamic environments, which reduces the influence of moving objects and reconstructs the background. The method starts by semantic segmentation and motion points detection, then removing feature points on moving objects. Meanwhile, a clean and accurate semantic map is produced, which contains semantic information maintenance part. Quantitative experiments using TUM RGB-D dataset are conducted. The results show that the absolute trajectory accuracy and real-time performance in dynamic scenes can be improved.

This paper designs a sliding mode self-pose attitude controller for a quadrotor unmanned aerial vehicle (UAV) system based on proportional integral observer, which also considers the problem of external disturbances. In this paper, firstly, considering the influence of external disturbance, a proportional integral observer is developed to contains the ratio of the estimated error and the integral loopobtain, and obtain both the estimation of the system state and unknown input. Then, the stability of the state estimation error of the proportional integral observer is proved. Thirdly, based on the results of the proportional integral observer, a sliding mode attitude controller is designed. This method is simple and easy to implement, which can improve the system robustness and eliminate the chattering phenomenon existing in the traditional sliding mode control. Finally, the simulation results show that the UAV system achieves better control effect and dynamic performance.

Aiming at the problem that the lunar rover path planning algorithm generally has slow convergence rate, falls into local optimal solution, neglects the mutual applicability of environment modeling technology and path planning algorithm, a comprehensive genetic algorithm based on virtual three-dimensional model is proposed. By setting the genetic factor unchanged, the terrain comprehensive cost function is added to set the fitness function. Using genetic algorithm and improved comprehensive genetic algorithm, after 100 simulation experiments, the improved comprehensive genetic algorithm has good search performance, fast convergence and high stability. The key words of this paper are as follows: path planning, terrain comprehensive cost function, genetic algorithm.

Unsupervised Learning based SLAM algorithm has lately drawn significant attention for its potential in label-free leaning ability and robustness to camera parameters and environmental variations. In order to achieve better robustness and accuracy, a multi-constraint learning model is proposed. In contrast to traditional geometry-based methods, multi-constraint unsupervised learning models optimize the photometric consistency over image sequences by warping one view into another; make the Network learning more geometrically information. A lot of experiments on the KITTI data set show that our model is superior to previous unsupervised methods and has comparable results with the supervised method.

This paper presents the system architecture of a driverless robot car, designed to participate in the Carolo-Cup, a competition regarding automated model vehicles. We describe an implementation that completed the different tasks in this competition on our EyeBot platform in detail. EyeBot has one RGB camera as well as three infrared distance sensors, and it's powered by Raspberry Pi 3B. We developed the lane detection algorithm using the OpenCV library and completed the traffic sign recognition task based on SVM which can be used offline. Experimental and simulation results recorded in real-time are also reported. The test result showed that our programs can run at high speed to achieve stable motion control in real time and complete all the tasks in the competition. The highlight of our work is that the whole system can run on a platform with limited computing resources.

In this paper we designs an automatic docking system to simulate the precision docking of the key components in aerospace assembly line. By using the robot to make the actual measurement of the experimental workpiece, we figure out the spatial coordinate system of the workpiece which correspond to the robot world coordinate system. This coordinate system is used to automatically control the robot to calculate the motion trajectory and complete the precision docking work. This system is characterized by the use of Siemens PLC to control the KUKA robot dynamically through the KUKA-TIA-Library interface, which can meet the requirements of flexibility and customization in aerospace industry.

In this paper, an inspection manipulator with a micro-camera at the end was designed for in-situ microscopic observation of samples on the material exposure platform. In order to achieve the positioning requirements, a novel position measurement method for the manipulator was proposed based on monocular microscope vision. First, a planar target was designed for feature detection. Next, the measurement task was performed in two steps respectively. On the image plane, a motion-based active calibration method was used to measure the translation. For the rotation around Z axis, the relationship between the image blur b and the rotation θ was fitted to estimate the rotation from images. In final, the proposed approach was tested in term of accuracy and robustness in ground experimental conditions. The focusing error of the piezoelectric motor was about 1μm and the obtained average alignment accuracies were less than 10μm in X, Y, and 0.002° around Z, respectively, which satisfied the positioning requirements of the in-situ observation.

The continuum robot, which is different from the traditional rigid robot, consists of a flexible structure, and becomes a research hotspot in the field of robots, because of its good flexibility and adaptability in the complex environment. According to the advantages of the continuum robot, it could be applied in medical, exploration and agricultural operations. This paper proposed a cable-driven dual-arm continuum robot with three-degree-of-freedom. The kinematics of posture parameters and driving parameters, as well as the statics, was established. And the dual-arm coordinated reachable work space was described according to the principle of constant curvature deformation, on the basis of which, the theoretical motion model of the continuum robot was simulated using MATLAB, and experiments of the posture of the dual-arm continuum robot was conducted on the prototype to verify the accuracy of the theoretical kinematics and statics model.