Table of contents

Intelligent robots show great demand for scientific research and industrialization in all walks of life. At present, robots have made breakthroughs in perception and motion control, but it will take some time for robots to fully serve autonomously. ICRAIC 2021 will focus on the development and integration of robot, digitalization, automation, and AI technologies, inviting well-known experts and scholars at home and abroad, as well as scientific researchers and industry practitioners in the field of intelligent control, to jointly discuss the cutting-edge technologies of robots and their new applications in different industries, and build a robot innovation ecosystem.

The International Conference on Robotics Automation and Intelligent Control (ICRAIC2021) is organized by Central South University, University of Chinese Academy of Sciences, and Nanjing Normal University and held virtually online due to the COVID-19 pandemic on November 26th-28th, 2021. ICRAIC2021 brought together researchers, developers, and practitioners worldwide in Robotics Automation and Intelligent Control.

ICRAIC2021 was structured around 2 tracks including Robotics Automation and Intelligent Control. Based on the two tracks, Prof. Hanxiong Li from City University of Hong Kong, Prof. Yong Wang from Central South University, Prof. Giuseppe Carbone from University of Calabria, Prof. Mustafa MISIR from Duke Kunshan University, and Prof. Xianhua Wu from Shanghai Maritime University made eloquent speeches. Meantime, additional programs included half a day workshop regarding Robotics Automation and Intelligent Control with 24 presentations.

List of Conference Committee Members are available in the pdf

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

• Type of peer review: Single Anonymous

• Conference submission management system: Morressier

• Number of submissions received: 201

• Number of submissions sent for review: 180

• Number of submissions accepted: 80

• Acceptance Rate (Submissions Accepted / Submissions Received × 100): 39.8

• Average number of reviews per paper: 3

• Total number of reviewers involved: 95

• Contact person for queries:

Name: Yong Wang

Email: ywcsu2@gmail.com

Affiliation: Central South University

Satellite antennas are widely used in satellite communications and now have become an indispensable part of satellite. It affects and restricts the performance and functions of the entire wireless communication system. Flexible pointing and low power consumption are required for antennas installed on satellites, which makes the design of the pointing mechanism of the satellite antenna more challenging. This paper focuses on the design and analysis of the pointing control mechanism to help improve the flexibility of satellite communication. An optimized structure consisting of spring hinge mechanism, manipulator arm and universal adjustment mechanism is proposed. The material selection and mechanical analysis are also made. Finally, the pointing angle is described based on the mathematical model of the pointing control mechanism.

The intersections are junctions of roads. Recognizing intersections based on LiDAR accurately and quickly is a key task for UGVs. In this paper, we propose an end-to-end real-time intersections recognition network (IRNet) with graph attention convolution based on graph classification and 3D LiDAR point cloud. Multiple evaluations on KITTI and Tunnel dataset demonstrate that our model performs better than other competitive models and meets the real-time recognition. We research the performance of our model under different number of input points, and certify that neither of two spatial transformation networks is effective for our model. Ablation experiments certify effectiveness of the proposed features skip connection.

The existing robot-world and hand-eye calibration algorithms always treated the robot pose as a deterministic value. Its uncertainty is not taken into account, which affects the calibration accuracy when the robot's position accuracy is low. In this paper, we proposed a factor graph approach to simultaneous robot-world and hand-eye calibration, which considers the measurement error of the robot joint encoder. And then, the nonlinear optimization of reprojection error minimization based on a product of exponentials (POE)-based model is implemented to solve the calibration problem based on lie groups. Through simulation and experiment, we validated that the proposed algorithm can get more accurate calibration results compared with the state algorithm.

In the traditional direct visual odometry, it is difficult to satisfy the photometric invariant assumption due to the influence of illumination changes in the real environment, which will lead to errors and drift. This paper proposes an improved direct visual odometry system, which combines luminosity and depth information. The algorithm proposed in this paper uses Kinect 2 to collect RGB images with the corresponding depth information, and selects points with large changes of gray gradient to construct a luminosity error function and uses the corresponding depth information to construct a depth error function. The two error functions are merged into one function and converted into the least squares function of the pose of camera, the Levenberg-Marquardt algorithm is used to solve the camera pose. Finally, the Graph optimization theory and the g2o library are used to optimize the initial pose. Experiments show that the algorithm can reduce the error to a certain extent and reduce the drift caused by illumination changes.

In this paper, a method to estimate the terrain based on footholds and least-squares is presented to make quadrupedal locomotion stable on the sloped terrains, with joint encoders, force sensors and inertial measurement unit (IMU) instead of visual sensors. This method which is tested successfully in simulation can predict two types of terrain, one for adjusting the robot pose and the other for adjusting the trajectory of swing phase. And to make full use of these new capabilities, general balance and locomotion controller is presented. The virtual model controller is embedded into architecture and it allows the robot to handle unexpected terrain. The quadrupedal locomotion in terrain with different inclination prove the robustness of this method.

Aiming at the problem that the influence degree of static and dynamic accuracy of EMA pitch servo mechanism is not clear, this paper analyzes the influence degree of static and dynamic error of EMA pitch servo mechanism based on high-precision pointing system. In terms of static error, based on the static error kinematics model, by establishing the static error probability model and solving the static error model, it is pointed out that the geometric error is the largest factor affecting the static kinematics error. In terms of the influence degree of dynamic error, based on the dynamic error kinematics model, through the establishment of the dynamic error separation process, and based on the test and mechanical experiment platform, it is pointed out that the general trend error is the largest factor affecting the dynamic kinematics error. Through the analysis of the influence degree of static and dynamic kinematic errors, the accuracy improvement direction of the EMA pitch servo mechanism is defined.

With high demand of automatic development of four-wheel alignment technology, this paper explores the core technology of automatic detection of four-wheel alignment instrument: intelligent identification and processing of rim target. According to the working process of 3D vision four-wheel aligner, starting from the target positioning, this paper proposes a method based on matching idea to realize the target recognition and positioning; Then the target area image is segmented to obtain the center coordinates on it, which provides a basis for the subsequent calculation of four-wheel positioning parameters. Simulation results demonstrate the effectiveness of our method.

The ocean contains huge valuable resources, due to the huge pressure in the deep water and the danger of diving, human development is very little. In this paper, we design a small underwater observation robot, which can replace human beings to dive into deep water, detect the deepwater area and sample the water quality. Following the principle of fish diving, the robot uses the piston bucket to pump water and drain water to change its gravity to achieve floating and diving. The main propeller at the tail provides the main driving force. The robot turns by propellers on both sides working at different speeds and propellers can change the thrust direction to provide assist force in the process of diving and floating. The structure design of the tumbler(the design of anti-roll) ensures the anti-interference ability of the robot to the wind and wave. In addition, the robot can be equipped with cameras, depth sensors, temperature and humidity sensors, and other sensors to achieve efficient underwater data sampling. The depth sensor, propeller, and pumping piston are used to control the hovering depth of the robot. The ROV can record good quality of video if provided with proper video graphic tool. The design of ROV which we have conceptualize, is handy and can be powered by 24V DC power which can be easily available at remote location. The experimental results verify the structural reliability of the underwater detection robot and provide a new mobile mechanism platform and a new idea for deep-sea exploration and scientific research.

The paper designs a Three-DOF assembly robot based on the planar quadrilateral mechanism. It is mainly used for the handling and assembly of parts, featuring small size, low cost and simple transmission principle. The robot combined with the characteristics of the planar quadrilateral mechanism, a detailed kinematic analysis is carried out and the forward and inverse solutions are obtained. The paper use MATLAB to analyze its end working space and motion trajectory according to the constraint equation. Finally, the motion simulation is carried out using ADAMS, which verifies the effectiveness of the design and provides a theoretical basis for the kinematics research and structural optimization of the assembly robot.

Robots often collide in the process of simulation soccer match. The collision will block the robot's own motion trajectory, which is not conducive to the robot's ball grasping and defense. Considering that the obstacles have the characteristics of dynamics, antagonism and uncertainty, the robot's motion trajectory, path selection and obstacle avoidance are very important. Taking ROS as the research platform, the study focuses on the path planning of simulated soccer robot. Meanwhile, we have applied some obstacle avoidance algorithms to the Middle Size Simulation League of soccer.

In this work, a real-time wide depth pose estimation method with laser scanning point cloud is proposed. Unlike other 6D pose estimation methods, the proposed method can maintain a robust and accurate tracking performance even when the target spacecraft has an obvious transformation in depth. To make full use of the known model, a pose estimation network based on CNNs is established to estimate the position. After the point cloud is translated according to the estimation result, the quaternion of the target is calculated independently. To evaluate the consequent of the network, we simulated the scanning point cloud with a wide depth range. The results show that the proposed method has strong robustness to the depth transformation and higher computational efficiency.

For purpose of efficiently extracting the fault detailed characteristics of rolling bearing vibration signals in the midst of heavy noise interference, the author combines the mutual information theory and proposes a singular spectrum decomposition effective component discrimination method based on the minimum mutual information criterion. First, the SSD algorithm is used to decompose the original vibration signal into a number of SSC; then the mutual information value between the original signal and each singular spectrum component is calculated separately, based on the nature of the bearing fault vibration signal and the mutual information theory for derivation and analysis, and the selection is determined the singular spectrum component with the smallest mutual information value among the original signal and each component is taken as the best component; finally, the best component selected based on the minimum mutual information value criterion is analysed by envelope spectrum to extract the fault characteristics of rolling bearing. Through the analysis of bearing fault simulation signal and the comparison with other indicators, it is shown that this set of experiments significantly extracts the characteristics of the vibration signals emitted by rolling bearings due to faults in noisy environments, thus revealing promising applications and development prospects.

Canopy cover estimation is widely applied to reflect crop status in agriculture research and management. In particular, an accurate CC estimation is beneficial for crop model calibration, providing high-accuracy observations. Recent solutions on CC are drawn by experimental regression or basic machine learning classifier because CC estimation can be treated as a wheat/non-wheat segmentation task. However, the appearance of hyperparameters in such machine learning algorithms impairs the segmentation performance. In this paper, by the means of UAV multispectral imagery, Bayesian optimization based Random Forest approach is selected to tune the uncertain hyperparameters accurately and robustly, providing a novel way on CC estimation. Experimental results collected in Yangling experiment field by the RedEdge camera on DJI M100 UAV are to evaluate the proposed method. Comparative studies show that the overall accuracy can reach up to 99.9%, promoting 0.2% in comparison with basic Random Forest. Therefore, integrating optimized Random Forest and UAV multispectral imagery can be applied in CC estimation at farmland scales.

Point cloud registration is one of the key issues in fields that need 3D scenes with global vision, including 3D scene reconstruction in robot technology, high-accuracy 3D map reconstruction in automatic driving, 3D reconstruction of real-time monitoring underground tunnel. Recently, point cloud registration methods based on deep neural networks have made great progress by learning from a large amount of well-aligned point cloud pairs. However, these large-scale labelled samples are usually difficult to obtain in the real-world applications such as visual simultaneous location and mapping of mobile robots which obtains unlabelled scene images in real time. To solve this problem, we propose a deep point cloud registration method based on the semi-supervised learning. Specifically, for the problem of less supervised data, we use Iterative Closest Point (ICP) algorithm to generate pseudo labels for the point cloud samples with unknown rigid transformations, effectively increasing the number of supervised data; in order to amend the pseudo labels, we design an alternating optimization algorithm to jointly learn the pseudo labels and deep point cloud registration model. The deep point cloud registration model and ICP are used alternately to continuously improve the precision of pseudo labels and the performance of the deep model. Experimental results show that this method can still obtain a reliable point cloud registration model even when just observing a small amount of point cloud samples with ground truth labels.

The key issue of meteorological temperature verification is how to operate itself without personnel deployment. This paper presents a meteorological temperature verification system based on robotic arm. Its goal is to realize the automatically operating by integrating robotic arm controller and verification software. The device level provides temperature verification instruments and the robotic arm. The communication function such as cyclic measurement of the resistance of temperature sensor, verification environmental data communication and temperature generator control can be realized by networking level. The interactive level ensures stable operation and fault location. Some key methods such as the system layout, the system workflow and the system integration method are also illustrated for system implementation.

Multi-agent system is an important branch of distributed artificial intelligence. As a new field of engineering control science, it has developed vigorously in recent years. From a practical point of view, most existing agents, such as unmanned vehicles and unmanned aerial vehicles, are powered by mobile power supply. Therefore, it is of practical significance to consider the battery state and explore how to improve the overall battery life when designing the multi-agent control strategy. In this paper, we first introduced a classical algorithm based on the consensus theory, and on this basis, the overtaking action in the formation is realized by changing the topology, so as to simulate the individuals (action individuals) who consume more energy in the formation; then we discussed how to improve the endurance of high-energy-consuming individuals in formation through average energy consumption in strategy design, and proposed an algorithm based on energy-aware. Finally, the MATLAB simulation results show that compared with the classical algorithm, the proposed algorithm can improve the endurance of formation.

The agricultural water-saving irrigation systems need to be adapted to local conditions, and irrigation emitters with different flow rates need to be designed to meet actual needs. In this study, machine learning approaches were used to predict the flow rate of the labyrinth emitters. The structural parameters of the emitter [number of channel units (N), channel unit length (L), channel width (W), tooth height (H) and channel depth (D)] and working pressure (P) were considered as independent variables. The applied machine learning models included k-nearest neighbor (KNN), multi-layer perceptron (MLP), support vector machine (SVM) and radial basis function artificial neural network (RBF-ANN). The accuracy of the machine learning model was evaluated by the mean square error (MSE) and coefficient of determination (R²). The results show that the MSE and R² of the RBF-ANN model were higher than the corresponding parameters of the other three models, indicating the RBF-ANN model can predict the flow rate of the labyrinth emitter more accurately. The research provides a reference for the rapid design of emitters with different flow rate. It is helpful to realize the automation of agricultural irrigation, and then realize the construction of a smart agricultural irrigation system.

This paper proposes a description of flexible spacecrafts' motion and rotation uniformly with dual quaternions. The dynamic equations are derived by using Lagrange equations. Recursive Newton-Euler (RNE) method is introduced to calculate the nonlinear terms in the equations. The simulation results demonstrate the coupling effects during the actuating speed switch of the solar panel.

Aiming at the problems of a single control mode, complex signal line, serious interference and poor system stability of a traditional weeding machine, a distributed control idea was proposed. Based on STM32 as the main control chip, each control unit realizes motion control, detection control, command control, and the terminal was displayed in a centralized manner. Based on ISO11783 protocol, according to the actual requirements of the weeding robot, the CAN bus communication protocol was improved and formulated to form a distributed control system. The system can completely realized all the actions of the weeding robot. 1. Motion controls: robot forwards, backwards, turning and weeding knife up or down; 2. Detection feedback: a power chassis climbing angle calculation and weeding knife speed self-adjustment, battery power, engine fuel capacity display; 3. Control command input functions. The test results had shown that the control system is stable, responsive solid, design specifications. It could meet the requirements of practical application. Meanwhile, the real-time accuracy and reliability of a CAN bus application in the agricultural production control system were verified.

Asteroid exploration missions demand more autonomous navigation system that are precise and in real-time. Visual-aided inertial navigation systems combine the advantages of autonomy of inertial navigation systems and accuracy of visual navigation systems. However, both increase the computational burden for the information processing. With the goal of decreasing the computational load, this paper proposes a low cost vision-aided inertial navigation method which can calculate increment position, velocity and attitude directly in inertial frame by pre-integrating IMU measurements. For visual information, specific combinations of the features are selected according to information entropy theory to reduce computation. Simulation results show that the standard deviations of the spacecraft's estimated position error is less than 1 m, the velocity less than 0.002 m/s and the attitude angle less than 0.005 deg, compatible with precision landing on asteroids.

To address the issue of complicated dynamics modeling and difficult solution of 3-RRR flexible parallel robot, we propose a method of building a flexible parallel robot dynamics model using screw bond graph. Firstly, the structure of the flexible parallel robot is analyzed; secondly, the multi-energy domain dynamics model of the robot is built using screw bond graph and vector bond graph considering the elastomeric transformation in the middle flexible linkage; finally, based on the model, the closed-loop control strategy for suppressing harmful vibration of the system, and the speed loop error compensation for different control parameters is also studied. Author establishes the vector bond and spin bond conversion junction type based on the power conservation transfer principle, which realizes the combination of rigid and flexible system in this paper. And we construct the dynamical model of multi-energy domain flexible parallel robot system.

Although Unmanned Aerial Vehicles (UAVs) are developed for military missions originally, they have been used widely in civil activities for several decades globally. In agricultural, UAVs have been developed as an efficient sprayer for pesticide application since 1987. UAV-based sprayers are popular for the prevention and control of pests and diseases in field crop in many countries recently. Some of the UAV-based sprayers are developed to be equipped with fruit tree mode aiming at solving droplet penetration in the inside and bottom part of the fruit tree canopy. In this study, a newly released UAV-based sprayer (i.e., T30) equipped fruit tree mode is chosen as spraying platform to optimize the spraying parameters for practical application. The flight velocity and application rate are the variables, while droplet coverage, density, size, and penetration are the observed metrics. Three treatments with different flight velocities (2 m s^-1 or 3 m s^-1) or application rates (60 L ha^-1 or 75 L ha^-1) are arranged to collect the droplets for assessment. Water Sensitive Papers (WSPs) are placed in the outside, bottom, and inside layers of the canopy to collect droplets. The results show that the treatment combined a flight velocity of 2 m s^-1 and an application rate of 60 L ha^-1 obtains the most droplets among all the variables based on the values of droplet coverage and density. The treatment with a flight velocity of 2 m s^-1 and an application rate of 75 L ha^-1 has the best penetration, while the treatment with a velocity of 2 m s^-1 and an application rate of 60 L ha^-1 takes the second place according to the percentages of droplet deposition in the three layers. Overall consideration of the total droplet distribution and penetration of the application parameters, a flight velocity of 2 m s^-1 and an application rate of 60 L ha^-1 are recommended to get an ideal droplet distribution in tree canopy when UAV-based sprayer T30 equipped with fruit tree mode flies at 1.6∼2 m above the citrus tree canopy.

Reconfigurable robots have excellent reconfigurability and environmental adaptability, but the configuration design method to select appropriate configuration with different environments has not been fully studied. Most of the existing work is based on the known global environment information, or only focus on the structure design. How to effectively obtain rich terrain information for subsequent configuration design is the subject of this paper. In our work, a comprehensive method of obtaining robotic configuration is proposed which considering geometric information and semantic information of unstructured terrain. And it provides new possibilities for reconfigurable robots to perform autonomous tasks in complex scenarios. Results of simulation experiments prove the availability of the mentioned method.

Using low-cost sensors to achieve high-precision positioning has been a hot research topic in autonomous driving technology, and smartphones not only have the advantages of low cost and easy installation, but also can provide high-precision navigation and positioning services. In this paper, we focus on the cycle-slip detection method based on the original Global Navigation Satellite System (GNSS) observations of smartphones. In response to the problem that the phase value of smartphone GNSS has serious cycle slip and the existence of frequent clock jumps will affect the cycle slip detection, a cycle slip detection method that integrates clock jumps and single and dual frequency observations is proposed. The experiments show that the method can effectively detect the continuous and as small as 0.5 week of cycle slip for the phase observations with the sampling rate of 1 Hz. The root mean square error (RMSE) in the L1 and L5 phase residuals after the cycle slip repair is 7.0 mm and 8.3 mm, respectively, with an accuracy improvement of 18.16% and 9.62%, which indicates that this method is effective and feasible.

Deployment of modern detection models is difficult for infrared target detection used in robot vision systems due to their heavy computational burden. To alleviate this situation, a simple but efficient channel pruning method is proposed for model acceleration. Specifically, a soft-gated module combined with batch normalization (SGBN) is designed as a standalone layer to substitute the standard batch normalization (BN) layer during training. The conversion between SGBN and BN is easy, and the training overhead introduced is almost negligible after replacement. By controlling the sparsity of the scaling factor in SGBN, unimportant channels with small output are blocked automatically and globally, which is simultaneous with model training. Removing these redundant channels no longer requires fine-tuning, thus significantly speeding up the pruning process. Experiments of pruning different detection models on the infrared dataset show the effectiveness of our method. For example, the parameters and FLOPs of pruned CetnerNet are reduced by 72.70% and 40.20%, respectively, without accuracy loss. The inference speed on the CPU is 12.01ms faster. Extended studies on the classification task also demonstrate its great potential when transferring to other applications.

Deep neural networks (DNNs) can be attacked by adversarial examples that are undetectable by humans. Generation-based approaches have recently gained popularity because they directly translate the input distribution to the distribution of adversarial instances, making them more effective and efficient. However, existing techniques are susceptible to overfitting on the substitute model, limiting the transferability of adversarial examples. In this paper, we introduce data augmentation into AdvGAN, called AdvGAN-DE, which can dramatically improve the transferability of adversarial examples. Experiments demonstrate that we enhance the success rate by 31.09% for defense models and 10.23% for typically trained models on average when compared to AdvGAN.

Unmanned aerial vehicle (UAV) low-altitude remote sensing image stitching is a new technology to promptly grasp the lodging situation of rice. The effect of image stitching depends on different application scenarios, so that it is necessary to explore low-altitude remote sensing image stitching algorithm suitable for rice lodging monitoring. The research adopts SIFT (Scale invariant feature transform) and SURF (Speeded up robust features) feature detection algorithms to conduct mosaic experiments based on drone images of a rice field in Dehui City, Jilin Province. The results demonstrate that the image stitching technology based on surf algorithm possesses better real-time performance, and the panorama obtained can well reflect the lodging condition of rice field. This research can provide technical reference for the actual lodging monitoring of rice field.

The built-in sensors of smartphone, such as cameras, accelerometers and gyroscopes, provide abundant environmental information for positioning. VIO (visual-inertial odometry) can be derived from the embedded camera and inertial sensors to provide the estimation of a continuous and smooth trajectory of the pedestrian. However, the trajectory will drift due to the accumulation of system errors. Currently, DTV (digital TV) signal has been verified to be promising for precise ranging in urban scenarios. Therefore, this paper investigates a method of fusing the VIO with the ranging information from DTV signal to improve the positioning accuracy. An Extended Kalman filter is presented to estimate pedestrian mobile state. Field tests have been carried out and the fusion results can effectively reduce trajectory drift errors and improves positioning performance.

SLAM (Simultaneous Localization And Mapping, SLAM) is the key technology for mobile robots to obtain information about their position and surrounding environment, but both laser radar and vision sensors have limitations in solving SLAM problems, for example: a single laser sensor can't do mapping effectively for hollowed-out obstacles, while a single vision sensor can't work well in areas without texture. Besides, the existing loop detection technology still has some problems such as large calculation resources and errors. To solve the above problems, an improved SLAM algorithm laser and vision fusion was presented in this paper. In this algorithm, a Gaussian statistical linear interpolation algorithm was used to interpolate the laser point cloud, and then the interpolated laser raster map and visual raster map are fused by the Bayesian method to obtain a denser and more accurate map. Furthermore, the histogram was used to improve the loop detection part to further improve the positioning accuracy. The improved loop detection algorithm was simulated and verified by running the TUM data set. On this basis, a mobile robot experiment system was built to do mapping in the laboratory scene. The experimental results show that the algorithm proposed in this paper can map the hollowed-out area accurately and effectively, it plays a significant role in improving the drawing effect of the robot, and can provide more accurate location information, which has good practicability.

With the continuous development of network, the increasing popularity of embedded technology, and the continuous improvement of mobile terminal computing ability, how to reduce the size of the model without affecting the accuracy as much as possible, and how to deploy the algorithm in embedded devices has become the current research hotspot. In this paper, we propose a new target detection algorithm called YOLOX-Mobile. Firstly, in order to further reduce the power consumption of embedded devices while maintaining the performance, we replaced the SiLU activation function in the original YOLOX with Mish activation function. Because SiLU activation functions occupy a certain amount of computing and storage resources, and the cost of calculating the Sigmoid type functions on mobile devices is much higher. Therefore, we use Mish functions that are smoother, non-monotonic, unbounded upper and lower bounds are used as activation functions to better meet the low power requirements of embedded devices. Secondly, we used Focal Loss as the Loss function of obj_output to achieve the balance of positive and negative samples, as well as hard-to-classify and easy-to-classify samples. Thirdly, we introduce the Involution operator and use it as the convolution kernel of 3×3. We validated the proposed algorithm on public dataset VOC2012 against the current mainstream YOLOX, YOLOX-M, and YOLOX-S algorithms. The mAP of our proposed algorithm is 78.22% and the detection speed is 55.26 FPS. Compared with the original YOlOX, YOLOX-M and YOLOX-S algorithms, the average FPS is improved by 1.99% and 4.13 FPS, and the average Params Size is reduced by 28.80%. Experimental results show that our proposed algorithm improves the accuracy and speed of detection on top of greatly reducing the network parameters and computation, making it a more suitable target detection algorithm for mobile devices.

This paper studies the design and control method of the cement silo wall cleaning robot system, analyzes the overall control requirements of the robot, and uses PLC as the control processor to design a cement silo wall cleaning robot control system, including action motor, pneumatic components and other components selection, which designs and completes working motors and sensors wiring schematic diagram. According to PLC input and output wiring diagram design, timing analysis and the human-computer interface of the upper computer, the design of the inner wall cleaning robot control system was completed, which can adapt to different circumstances, and has the characteristics of automatic cleaning, continuous operation and efficient completion. On this basis, aiming at the functional needs of the robot cleaning the inner wall of cement silo, cleaning path planning and control has been designed with better conditions.

This paper proposes a calibration method of RGB-D camera, especially its depth camera. First, use a checkerboard calibration board under auxiliary Infrared light source to collect calibration images. Then, the internal and external parameters of the depth camera are calculated by Zhang's calibration method, which improves the accuracy of the internal parameter. Next, the depth correction model is proposed to directly calibrate the distortion of the depth image, which is more intuitive and faster than the disparity distortion correction model. This method is simple, high-precision, and suitable for most depth cameras.

Time series forecasting play an important role in many applications, and combining neural networks to forecast time series features can uncover many potential application situations. For example, forecasting traffic flow time series data is important for urban traffic planning and traffic management. Accurate forecasting of traffic flow provides the basis for road traffic control, which in turn improves traffic efficiency and reduces congestion. In this paper, STSeNN is proposed, which combines graph neural network and TCN to dynamically capture the spatial correlation, temporal correlation and semantic correlation of data. Experiments show that the prediction performance of STSeNN on the traffic flow datasets is more accurate compared with existing methods.

Neural Network is an excellent methodology for predicting lithium battery state of health (SOH). However, if the data amount is insufficient, the neural network will be overfitted, which decreass the prediction accuracy of SOH. To solve this issue, a data augmentation method based on random noise superposition is proposed. The original dataset is expanded in this approach, which enhances the neural network's generalization ability. Moreover, random noises simulate capacity regeneration, capacity dips and sensor errors during the actual operation of lithium batteries, which also improves the adaptive and robustness of the SOH prediction method. The proposed method is validated on mainstream neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) neural networks. In terms of the results, the proposed data augmentation method effectively improves the neural network generalization ability and lithium battery SOH prediction accuracy.

The network model using deep learning have been widely used in the sphereof visual object tracking for the past few years. The Siamese network can utilize the model based on deep learning to achieve a balance between the tracking accuracy and speed in the visual object tracking. This work mainly introduces the development process of the visual target tracking field and traditional target tracking algorithms. It focuses on the Siamese network structure and the improved the Siamese algorithm, and compares tracking results of related algorithms. Aiming at the deficiencies of existing Siamese object tracking algorithms, the future development trend is prospected.

With the rapid development of powered transtibial prostheses, the control of motorized actuators is a research hotspot. This study designs a double-loop impedance controller to improve the performance of the traditional controller. Including time-varying parameters, the designed controller is implemented on a novel elastic actuator. Under different loads of users' weights, simulation tests are conducted in a working condition of walking, and the results show that the designed control system has outstanding performances in the control of an entire gait cycle. The error between output and reference of ankle joint position during walking is less than 0.013rad in each sample time. A position disturbance test demonstrates that the control also has a high tolerance of any position disturbance at the ankle joint. The double-loop impedance controller can replace traditional finite-state impedance control in the field of the powered transtibial prosthesis and has the potential to be applied in other relative fields.

The purpose of image inpainting is to automatically repair damaged areas using relevant information from preserved areas. Recent years, with the advancement of deep learning, significantly improved performance of image drawing has been achieved. In this paper, we are committed to reviewing the key techniques for automating image inpainting research. The article briefly describes conventional methods while focusing on deep learning-based inpainting methods, including model classification, strengths and weaknesses, range of usage, and performance comparison. Finally, the current issues and tendencies of image inpainting automation are discussed and predicted.

Tasks in real-time systems have strict time requirements and need to be completed before their deadlines. Whenever the real-time task set changes, the real-time system reruns the schedulability analysis method to determine whether the tasks in the new task set can be completed before their deadlines. Since new tasks can not be executed until the execution of the schedulability analysis method is completed, the execution efficiency of the schedulability analysis method directly affects the actual performance of the real-time system. Response time analysis (RTA) is a typical method for schedulability analysis. It calculates the response times of tasks and tests the schedulability of tasks by comparing the calculation results and the relative deadlines of tasks. We found that the execution efficiency of existing RTAs can be improved by reasonably sorting the calculation order of the tasks. A task sequencing strategy is proposed to improve the execution efficiency of existing RTAs.

Virtual Synchronous Generator (VSG) control strategy can provide inertia and damping support for the power system by simulating the operating characteristics of synchronous generator, and improve the frequency response of the power system. First, analyse the influence of various parameters on the output characteristics of the system by establishing a mathematical model of VSG. Secondly, to fully utilize the flexible and controllable characteristics of VSG control strategy, the basic principle of VSG parameter adaptation is obtained through the analysis of the angular frequency oscillation characteristics of synchronous generators, and an adaptive control strategy of virtual inertia and virtual damping parameter is proposed, which can better track changes in frequency, and set the action threshold for adaptive control. Finally, compare the frequency and active power response between the traditional VSG control strategy and the adaptive one and verify the effectiveness of the proposed control strategy.

Hyperspectral image classification technology is a basic work in the application of hyperspectral images. In recent years, with the innovation and development of hyperspectral image classification technology, the method and performance of hyperspectral image classification based on joint spatial-spectral features have made breakthroughs, and have gradually become the focus of researchers. In order to further promote the development of the spatial-spectral feature union class method and improve the classification accuracy of hyperspectral images, this paper summarizes the commonly used spatial-spectral union algorithms. Firstly, the introduction briefly outlines the background and research status of this field. Some common problems in the process of hyperspectral image classification are listed. Finally, some current hyperspectral image classification methods based on joint spatial-spectral features are introduced. The main roles and existing problems of spatial-spectral joint features in the field of hyperspectral image classification are summarized in detail, and the future research directions are prospected.

This paper presents a temperature control framework of Diesel Oxidation Catalyst (DOC) of after-treatment system (ATS) which aims to increase the control precise of DOC temperature under different exhaust flow rates. To start with, a control-oriented DOC temperature model is established. Considering the sensitivity of DOC to different exhaust flow rate, a fuzzy adaptive PID temperature control system framework under the Smith predictor is designed. Finally, the controller is verified in the Matlab/Simulink. The results show that when tracking the desired temperature, the overshoot is reduced and the response speed is faster at different exhaust flow rates compared with the traditional PID control system.

Based on large-scale storage and networked computing resources, cloud computing technology provides a strong computing platform which could support for massive data storage and big data intelligent application in the different field. Space-based information infrastructure has matured over the years, but most of the satellite systems are independent, which cause the resources of the satellite computing platform are isolated from each other. In this case, storage and computing resources cannot be utilized fully, which also cannot effectively support space-based intelligent application platform[1]. In this paper, a space-based embedded cloud computing platform is designed by combining the application requirements of space-based intelligent computing and ground-based cloud computing system, which includes sharing and using large-scale computing resources, concurrent monitoring and management of multi-tasks, fault-tolerant migration of tasks, and distributed storage retrieval. The results of this paper are committed to building an intelligent information network computing and storage platform, which could support for space-based intelligent infrastructure application.

A novel robust adaptive control is proposed for target detection and recognition in transmission line inspection, this paper proposes a quadrotor inspection system based on robust adaptive control of multi-sensor fusion. Robust adaptive control ensures the stability and anti-disturbance ability of the quadrotor during the flight. It is measurable for the distance between the quadrotor and cables by laser sensor and ultrasonic sensor. It can be used to detect and identify the target objects on cable lines by OpenMV. According to the proposed Lyapunov function, the asymptotically stable conditions is obtained, in which all trajectories converge in finite time to the desired location. Finally, the effectiveness of inspections is validated experimentally, and the system has strong anti-interference ability.

The quality of channel estimation (CE) is critical to the performence of wireless communication systems. In addition to the conventional model based CE methods appeared in the past few decades, deep learning (DL) based CE methods, which is introduced to learn the statistical characteristics of wireless channels, has emerged as a promising methods in recently years. In this survey, different DL models applied to solve CE problems are discussed firstly, then DL-based CE problems in different wireless communication systems are investigated.

The accuracy of pig target detection is not ideal due to insufficient light in a real pig farm environment. An improved enhanced network helps to increase the accuracy for pig target detection. The ResNet-Attention-RetinexNet algorithm (RA-RetinexNet) is proposed to solve the problems that YOLO V4 has low accuracy in pig image detection under low light and Mosaic data enhancement method cannot improve the brightness of low light image. In this model, we build residual connection and add attention mechanism, decreasing data loss of RetinexNet and enhancing the brightness information of images. The experimental results show that the model achieves better performance for pig target detection under low light.

Planetary rover plays a significant role in the lunar and Mars exploration missions and wheel's move status such as sinkage and slip ratio are the critical Factor to estimate the moving ability of the Mars rover. A method is proposed to estimate wheel's sinkage and skid angle based on terramechanic model and estimate slip ratio based on rover's dynamic model using Extended Kalman Filter algorithm. The proposed method can avoid the errors caused by Visual Odometry and improves the estimation accuracy of rover state through the fusion with other sensors' data. The effectiveness of the estimation method for planetary rovers is verified by comparing with the results obtained from dynamic simulation of Mars rovers.

In this paper, we apply the Long Short-Term Memory (LSTM) network to short-term load forecasting, and use the TensorFlow deep learning framework to build a Particle Swarm Optimization (PSO) model to optimize the parameters of the LSTM. Optimization (PSO) model to optimize the parameters of LSTM. In this paper, we use the meteorological data and historical load data of a certain place as the input of LSTM before and after optimization, and compare the model with the BP Neural Network before and after optimization, and the results show that the PSO-LSTM model has higher reliability and prediction accuracy.

Although federated learning can break the data island and enhance privacy, it is not absolutely secure. External eavesdroppers can intercept the model when clients upload their models to the server, which might be used to infer the data information. A natural solution is to add noise to the uploaded model. However, this solution has its drawback, because the added noise cannot be eliminated when the models are aggregated at the server, which leads to poor performance. In order to tackle this problem, we propose a simple but effective algorithm, FedNoise. In the vanilla federated learning algorithm, a gradually decreasing learning rate is applied at the client side. While in our algorithm, we divide the parameter update into client-side and server-side. The stochastic gradient descent update with constant learning rate is performed at the client side and a gradually decreasing learning rate is performed at the server. Our algorithm can not only ensure the data security, but also maintain comparable system performance. Simulation results validate our conclusion.

Affected by expert experience and individual differences in sows, artificial identification of sows in estrus is prone to various problems leading to low efficiency. A reliable audio classifier helps to identify the sows for estrus. In this work we propose a novel approach for estrus sound recognition using a fusion of two representative features as inputs and Convolutional Neural Networks (CNN) as training model. We demonstrate that the fusion feature generalizes better on the CNN model with training data under a real farm environment and gives remarkably higher test performance. Our model is also compared with some works that are published recently and achieve state-of-the-art performance.

This paper focuses on the research of object detection technology in Smart Campus scene. Firstly, an object detection algorithm is constructed based on Faster R-CNN two-stage detector, which is trained on COCO2017 and VOC0712 public datasets, and the performance of the trained detector is evaluated. On this basis, the actual image data is collected from the campus scene, and divided into different groups according to the shooting angle, lighting conditions, occlusion, etc. After using the algorithm to detect, the results are analyzed systematically. In general, occlusion between objects is more likely to lead to unsatisfactory detection results. In the future, we still need to find a non-maximum suppression method which can effectively distinguish the same object repeated detection and occlusion between objects.

Value iterative networks (VINs), as a differentiable path planning method, taking environmental images as input, can solve path planning problems in new and unseen environments with powerful generalization capabilities. But the planning success rate decreases rapidly as the planning space scales. We propose the potential field augmented VIN by replacing the environmental map with the potential map as input and implementing this process in the form of dilated convolutions, giving more a priori information to the subsequent planning computation while hardly increasing the computational cost. Experiments on 2D grid-world show that the improved method has higher path planning success rates and smaller difference of predicted paths from shortest paths, as well as smaller performance degradation with increasing map size.

On the visual side of computer science, image data is very important in the training of neural network models. Sufficient training data can alleviate the over-fitting problem of the model during training and help the model obtain the optimal solution. However, in many computer vision assignments, it is not easy and costly to obtain sufficient training samples. Therefore, image augmentation has become a commonly used method to increase training samples. Generative Adversarial Network (GAN) is a generative method of machine learning that can generate realistic images and provide a new solution for image augmentation. This article first introduces image augmentation and its commonly used four types of methods. Secondly, the basic principles of GAN and its direct and integrated methods in image augmentation are introduced, and the typical methods used to calculate whether the images from the networks meets the requirements; then the research status of GAN in image augmentation is analyzed. Finally, the problems and development trends of GAN model in image augmentation are summarized and prospected.

In the engineering implementation of gyrowheel, the influence of orthogonal support non-coplanar is inevitable. However, when the gyrowheel is used for satellite inertial angular velocity measurement, this non-ideal factor will inevitably have an adverse impact on the accuracy of angular velocity measurement. Therefore, the dynamic model of gyrowheel with non-coplanar orthogonal support is established, and the influence of the non-coplanarity support on the vibration frequency is analysed. Finally, the correctness of this research is verified by simulation.

Currently, indoor location-based services (ILBSs) have increasing requirements in people's daily life. In the meanwhile, the wearable devices are becoming more popular. In this paper, we studied a wearable system for indoor localization mainly based on INS/UWB. In order to achieve high-precision, stable, and continuous positioning, a sensor fusion method with anomaly detection is proposed. In the method, the sensor fusion method is derived from Bayesian estimation and a particle filter is developed to solve the nonlinearity problem and non-Gaussian errors for indoor positioning. In addition, the anomaly detection eliminates effects of NLoS and multipath effects significantly with the Mahalanobis distance. Two field experiments are conducted, and the results demonstrate that the 90% error of the proposed adaptive particle filter is 0.53 m, which is a 40% decrease compared with the PDR-only and UWB-only and classic PF, indicating better robustness and stability.

Based on the special differential game model of evader and pursuers, in this paper, we propose an optimal escape and containment strategies of evader and pursuers, and gives the escape boundary conditions of evader. Through the training of DDPG and A3C algorithms, the evader have the ability to escape autonomously, and the evaluation system is established with the reward value, the number of rounds, the success rate of escape and the ratio of escape speed as the main indicators. Taking different pursuer-evader speed ratios under DDPG and A3C as key indicators, a quantitative evaluation system was established, and the evaluation results verify the effectiveness of the proposed approach.

Most current knowledge base question answering models mainly use RNN and its various derivative versions such as BiLSTM to model the problem, which limits Parallel computing capabilities of the model. In response to this problem, we try to use TransformerEncoder instead of BiLSTM to model and encode the problem, and hope to improve the parallel computing efficiency of the model. At the same time, to solve the problem of insufficient relative position information obtained by using absolute position coding in TransformerEncoder, it is proposed to use relative position coding instead of absolute position coding. According to the experimental results, our model effectively reduces a certain amount of training time and has achieved certain results on the WebQuestions benchmark data set.

Preoperative Magnetic Resonance Image (MRI) brain tumor diagnosis is an effective technical approach. To accurately segment tumor regions, we propose a novel brain tumor segmentation method based on improved Spatial Attention mechanism and Multi-path neural network (SAMPU-Net). Firstly, we propose a multipath input method to extract feature information of different scales by using convolution kernels of different sizes, so as to fully extract MRI feature information. Secondly, we improve the spatial attention mechanism by adding convolution layer of pyramid structure to it to obtain the features of different receptive fields. In the convolution layer of this pyramid structure, the larger the convolution kernel is, the more global features will be extracted; conversely, the smaller the convolution kernel is, the more local features will be extracted. Thirdly, we use more multi-mode MRI information to segment the brain tumor images. In practical application, due to the fuzzy tumor regions in some MRIs, we use the method of restricted contrast adaptive histogram equalization to perform local enhancement of images. The proposed model and several other mainstream segmentation methods are trained and tested on the BraTS2019 public dataset. Experimental results indicate that using our method, the Dice coefficient of tumor core and tumor enhancement region is increased by 2.4% and 1.3% respectively, and our proposed method has better segmentation effect than other methods.

This article proposes a method of model predictive control, which combine the excellent data-driven optimization ability of reinforcement learning and model predictive control to design the controller. Different from the off-line design of MPC, reinforcement learning is based on the adaptation of on-line data to achieve the purpose of control strategy optimization. The reinforcement learning-based model predictive control can improve the control performance effectively. And the numerical simulations are given to demonstrate the effectiveness of the proposed approach.

Multi-agent collaboration is the core task of multi-agent system (MAS) research. In order to explore a kind of multi-agent collaboration way with dynamic environment, this paper proposes a method based on fuzzy learning and applies it to RoboCup2D. In the RoboCup2D simulation competition, eleven players need to cooperate to win, so score becomes extremely critical. Therefore, this paper focus on shooting technique based on fuzzy control. In order to increase the scoring rate, We speculate through observation and testing that finding the correct shooting position and angle is the key. For solving the problem, this paper firstly design fuzzy controller for selecting the shooting path; and then, this paper purpose fuzzy neural network for optimizing shooting time; At last, this paper do lots of experiments to verify its effectiveness and find it does work on increasing the scoring rate. After optimization on the fuzzy control, we are fortunate to have won the national second prize in 2021 RoboCup China Open, and the goal accuracy has increased by 21.36%.

In this paper, an optimization algorithm of forward-reverse cyclic assembly network under uncertainty is studied. A chance constrained algorithm based on robust approximation is adopted, in which the size of uncertain set is used to describe the violation probability of constraint. Further, some parameters are set to describe the confidence probability of the model. By implementing the proposed algorithm, we can reduce the manufacturing cost of the final forward-reverse cycle assembly network by 3.17% while ensuring a certain confidence probability, and relax the lower bound of the probability of the model, which makes the model more adaptable. By calculating the scheduling data and comparing with flexible robust optimization, the effectiveness of the algorithm is proved.

The traditional constant water supply system has problems, low operation efficiency, and the strong nonlinear and large time delay of the control system. It did not meet the needs of water supply in remote and desert areas. To solve these problems, the PV constant voltage water supply system is designed to optimize the Proportion Integration Differentiation (PID) controller method based on the particle swarm optimization (PSO) algorithm. The programmable logic controller (PLC) and frequency converter are used as the core control mechanism of the system, the water pump motor as an actuator, the pressure sensor and the level sensor be used as a feedback mechanism, Photovoltaic power generation equipment is used in power supply agencies. In this paper, the objective utilizes the PSO algorithm to optimize the parameters of the PID controller, and the principle and control process of the PSO-PID algorithm are analyzed. Simulation results by MATLAB verify that the PSO-PID algorithm has better dynamic and steady-state performance than the traditional PID algorithm. Finally, the practicability and effectiveness of the water supply system are proved by the PLC of the OMRON-CP1H series and the touch screen of the WEINVIEW series.

This paper investigates the problems of distributed cooperative cruise control and train running comfort for high-speed trains with input constraints under moving block mode. Firstly, inspired by the mode of hitting a hard wall for moving block, a distributed cooperative cruise control algorithm based on the position and speed state information of adjacent trains is proposed. Under this control algorithm, all trains can track the given speed and realize speed cruise. In this algorithm, the relationship between speed and displacement is used to improve the position term, so as to achieve the minimum safe distance between trains by adjusting the speed. Secondly, aiming at the problem of the excessive acceleration of each train in the moving block mode, the hyperbolic tangent function is introduced to limit the acceleration of each train in an appropriate range by limiting the control input. Finally, the stability of the closed-loop system in error state is proved by using Lyapunov stability theorem, and the feasibility of the algorithm is verified by simulation.

Pick and place (PAP) skill learning is a fundamental ability of intelligent robots, such as home service robot. Due to the NP-hard nature of the PAP problem, it takes a long time for an intelligent robot to learn the PAP skill based on current methods. In order to improve the learning efficiency of robot PAP skills, this paper proposes a Soft-DDQN-based PAP skill learning method. Firstly, the Soft-DDQN is proposed by introducing maximum entropy into robot DDQN framework, and the learning goal of Soft-DDQN is to maximize reward and information entropy. Secondly, PAP problem is modelled as a discrete form and Soft-DDQN is applied to solve the PAP problem. Finally, in order to verify the efficiency of Soft-DDQN-based PAP skill learning, comparisons have been given from two standard perspectives and shown that Soft-DDQN improves efficiency of PAP skill learning evidently.

The characteristics of aircraft maintenance training are that the teaching hardware resources are few, the maintenance training items are many, and the students are difficult to master. To solve this problem, we developed a simulator for aircraft maintenance training by studying the key technologies of maintenance training devices (MTD). Based on the classic control theory to design the autopilot control law, using big data from aircraft operation and scheduled maintenance to achieve high simulation MTD, to solve the difference between traditional MTD teaching and actual aircraft maintenance makes CMTD more effective and vital.

There are many kinds of inverse kinematics solutions for robots. Deep reinforcement learning can make the robot spend a short time to find the optimal inverse kinematics solution. Aiming at the problem of sparse rewards in the process of deep reinforcement learning, this paper proposes an improved PPO algorithm. Firstly, built a simulation environment for the operation of the robotic arm. Secondly, use a convolutional neural network to process the data read by the camera of the robotic arm, obtaining a network about Actor and Critic. Thirdly, based on the principle of inverse kinematics of the robotic arm and the reward mechanism in deep reinforcement learning, design a hierarchical reward function containing motion accuracy to promote the convergence of the PPO algorithm. Finally, compare the improved PPO algorithm with the traditional PPO algorithm. The results show that the improved PPO algorithm has improved both the convergence speed and the operating accuracy.

At present, the research on rectangular grasp strategy is generally based on object detection algorithm, which limits the improvement of model accuracy and generalization performance. This paper studies the semantic segmentation model based on residual network, and uses it to generate grasp strategies. The improved algorithm model not only achieves excellent rectangular grasping strategy prediction, but also has good generalization performance.

An improved Squirrel Search Algorithm (ISSA) is proposed to overcome the shortcoming of Support Vector Machine (SVM) parameters is difficult to select. Firstly, an adaptive probability formula is used to balance the exploration and foraging ability of SSA, so as to improve the optimization performance of SSA. In order to test the optimization performance of ISSA, two test functions are used. And results show that ISSA has better optimization performance than SSA. Finally, based on ISSA-SVM and SVM, and 3-fold cross-Validation is used, the fault diagnosis of DGA data is carried out. The diagnostic results show that ISSA-SVM has the highest diagnosis performance.

The fuzzy broad learning system (FBLS) is a novel, neuro-fuzzy model. Different from other neuro-fuzzy models with low efficiency, FBLS can obtain better performance using less computation time. However, the clustering-based fuzzy rule generation approach makes the performance of FBLS limited. Meanwhile, it is unknown how the enhancement layers from FBLS contribute to the model performance, which hinders the further extension of the model structure. To solve these problems, we propose a sparse enhancement fuzzy broad learning system (SEFBLS). It uses only a sparse set of enhancement nodes to replace the original enhancement node groups. To obtain a better representation, the designed principal component-based sparse autoencoder is used for feature reconstruction and information removal. In addition, to explore the optimal model structure and performance, multiple clustering methods (fuzzy and non-fuzzy) are used to improve SEFBLS. The results on 10 UCI classification datasets show that the proposed SEFBLS obtains competitive accuracy using fewer fuzzy rules.

With the miniaturization of electronic components, printed circuit boards (PCB) as a important component of them has become more and more complex and exquisite. Different defects in PCB can be found in modern industrial manufacture, such as lack of weld, lacking components, continuous welding, and the color ring resistance, resulting in low production. Accordingly, it is necessary to achieve the high rate of precision and efficiency when defecting during PCBs industrial production process. Considering that existing detection of PCB defects have a low efficiency, we propose an improved model YOPCB defecting PCB surface defects. Our improved model can achieve the precision of 97.6% mAP, which means our method is more efficient.

The brake system has the characteristics of multi-component, multi working conditions and complex degradation process, which brings great challenges to its health condition assessment. As it is difficult for a single brake agent to make a comprehensive and accurate health condition assessment, a health condition assessment model based on multi-agent federated learning is proposed in this paper. The different agents train their brake health data under different working conditions and states, which ensures the accuracy of health condition assessment and the safety of data of each agent. Aiming at the problem that it is difficult to determine the credibility of agent data in the process of federated learning, a credibility of agent data scheme based on evidence theory is proposed, which not only reduces the cost of calculation and communication, but also further improves the accuracy of health condition assessment. Simulation results show that the scheme not only has higher prediction accuracy, but also can ensure the security of agent data compared with the conventional centralized training scheme.

The damage stability of ship is a key performance metric in ship design. The challenge of damage stability optimization is a complex calculation process with multi scene dynamic inflow simulation relying on the cooperation of ship design software such as NAPA. Our work is to establish a collaborative optimization framework to effectively coordinate reinforcement learning (RL) with particle swarm optimization (PSO) and NAPA. The collaboration of RL and PSO realizes the update direction selection of watertight bulkhead position scheme. The collaboration of PSO and Napa promotes the iterative process of watertight bulkhead position and damage stability value A. The experimental results show that compared with the traditional damage stability optimization method, the collaborative optimization method improves the damage stability by 2.36% and reduces the calculation time significantly.

Intelligent production scheduling is an important part of intelligent production, and its production time and intensity are reasonably arranged, which can fully improve production efficiency. This article provides an intelligent model for engine manufacturing and scheduling. After the model decomposes the order into processes, the DSSM algorithm solves the semantic similarity of the text in the production content of the process, and generates a semantic vector to convert unstructured data into structured data. Based on XGBoost algorithm, the required production line is marked to establish the association between the production content and the production line, and realize automatic matching. After classification, the average working hours and other information are estimated according to the established logic. Finally, schedule production based on the principle of maximum utilization of the production line. Through testing and evaluation, the scheduling results of this smart scheduling model can be completed in just a few seconds, which saves time and costs in the scheduling process.

Grid-connected inverters incorporate distributed energy sources such as photovoltaics and wind energy into the micro-grid. Due to the volatility of distributed energy generation and the presence of a large number of non-linear loads in the micro-grid, harmonic wave seriously affect power quality. The grid-connected inverter with active filtering function is formed by adding the harmonic wave suppression function to the grid-tie inverter, which not only realizes the inverter grid connection but also achieves the purpose of harmonic wave compensation. The traditional model predictive control (MPC) has the characteristics of strong robustness and fast dynamic response, but its high switching frequency causes serious heating of the inverter. In response to the above problems, combined with hysteresis control theory, a model control strategy based on hysteresis is proposed in this paper, which improves the accuracy of grid-connected current of the grid-connected inverter with active filter function and inhibits the harmonic component of the system, verifying the effectiveness of the strategy.

In this study, a novel multi-objective antlion optimization (MOALO) algorithm-based impact load prediction model is proposed to address forecasting problem in the area with a lot of impact load. To reduce the impact of glitch in impact load data, ensemble empirical mode decomposition(EEMD) is applied to decompose the primitive load data into a set of sub-layers. Then, a novel MOALO-based extreme learning machine (ELM) forecasting model is put forward to make short-term prediction by using the decomposed sub-series. Finally, superimpose the prediction results. According to case study, the proposed EEMD-MOALO-ELM model has the best prediction accuracy and stability.

Energy is an important material that promotes social development and maintains human life. Clean energy has become a scientific research theme of open world which appeal to many scientists to research it. This paper inspired by the experience of sunflower growth. So we proposed a mathematical model for automatic adjustment of the maximum luminous flux for a limited solar panel, which can help to adjust the best light position. The physical model added to the light detector in the practical application, real-time comparison of the difference in illumination at different positions to adjust the direction of solar panel, thereby enhancing the utilization of light energy. In addition, in order to save space and prevent severe weather attacks, we designed a autofolding circle disc and retractable rectangular solar panels. The intelligent system judgement and decision making based on the electrical signals fed back by the wind sensor or the vibration sensor. When the wind speed reaches the threshold, it can be folded or contracted automatically, reducing manual maintenance. The fully automatic solar panels are designed in this paper, the research results show that theis model has a high utilization, strong sensitivity, safety and durability, and easy control. This is also a process of theoretical exploration of interdisciplinary services for the energy system.

This paper presents a penetration depth prediction model based on data fusion. The parameters of the penetration analysis are divided into different evaluation spaces, and then empirical algorithms are evaluated and the better algorithm is selected in each evaluation space. A large number of simulation data is generated to solve the problem of lack of experimental data. Two BP neural network prediction models are built based on experiment data and simulation data, respectively, and the genetic algorithm is used for parameter optimization. Finally, the attention mechanism is used to fuse the two models to generate the final dimensionless penetration depth. The experiment results show that the data fusion model has good prediction accuracy both in the whole parameter space and each evaluation space.

Honeypot is a proactive defense technology introduced by defenders. Through the honeypots, defenders can deceive attackers to illegally take advantage of the honeypots and capture and analyze the attack behaviors to understand the attack tools and methods. To build honeypots, defenders first imitate vulnerable systems to entice the attacker to attack, then deploy monitoring systems that is responsible for monitoring and recording the attacker's behavior. It is of concern that monitoring system is the key to determine the performance of honeypots, because obtaining attackers' behavior is the main purpose of deploying honeypots, and monitoring system's performance determines whether attackers' behavior can be accurately and comprehensively recorded. In this paper, we introduce a novel TaintDroid based honeypot monitoring system for embedded device. This system uses TaintDroid to mark the attackers who hack into the honeypot, monitors the behavior of the marked attackers and then records. Moreover, we tested the feasibility of this system by building a monitoring system based on TaintDroid.

Most Android mobile anti-virus software in the industry is checked at the application level, and users familiar with the Android operating system are well aware that the use of virtual clicks, function execution, or shell commands can force the application to stop, which poses a threat to the real-time monitoring of anti-virus software. Moreover, the current mainstream anti-virus software in the industry can only let users manually uninstall or deactivate malicious apps when detected, which also makes the anti-virus software in Android mobile lose the ability of mobile anti-virus software to remove or delete viruses and Trojans automatically. To solve the problems above, in this paper, we train a mobile anti-virus model based on Resnet50 and proposes an Android mobile anti-virus method using remote thread injection - overriding the execution of malicious code by RTI means such as hook API, nulling related functions, rewriting related classes or functions to preserve the app as much as possible. In contrast, The model can identify malicious code with the highest accuracy. The model's recognition accuracy is up to 98.14%, and the malicious code blocking rate is up to 99.70% after recognition.

Evaporation duct is a kind of special atmospheric stratification that frequently appears on the sea surface, which has an important influence on the propagation and attenuation of electromagnetic waves, and is an important factor affecting the efficiency of marine radars and communication equipment. After the development in more than half a century, evaporation duct height can be obtained by direct detection, theoretical model, inversion and machine learning. Machine learning can explore the hidden laws of data efficiently and has the potential to surpass the traditional theoretical model. In this paper, the Machine Learning methods in evaporation duct research are shown and prospects of machine learning methods in evaporation duct research are given.