Adaptive identification method of new energy grid operation risk based on linear decision function

The current conventional risk identification method of new energy grid operation mainly achieves risk identification by mining equipment status information, which leads to poor identification effect due to the lack of effective extraction of abnormal data features. In this regard, a linear decision function-based adaptive risk identification method for new energy grid operation is proposed. The data flow model is constructed by combining the linear decision function, and the features of interval abnormal data and fluctuating abnormal data are extracted. A sliding window model is constructed, and the unsupervised model is used to realize the effective update of new energy grid operation data. In the experiments, the proposed method is verified for recognition accuracy. The experimental results show that when the proposed method is used to identify the risk of grid operation, the effective data mis-deletion rate of the algorithm is low and has a more desirable identification effect.


Introduction
At present, the risk identification of new energy grid operation mainly relies on the parameters of electrical equipment to determine the operation status by checking whether the parameters of electrical equipment are in a normal fluctuation state to determine whether the equipment is malfunctioning at the moment [1].The acquisition of equipment parameters can be made through electrical sensing equipment.The process is usually affected by the acquisition environment and acquisition means, resulting in the acquisition of data containing a large number of noises, which also brings greater difficulties to the analysis of equipment parameters [2].The conventional method for the above problem is optimized by combining data noise reduction techniques to reduce the noise of the collected electrical equipment data and then extract the equipment data features.This method can achieve effective identification of grid operation risk to a certain extent, but there are also certain limitations [3].For example, the risk level of the equipment operation status classified by the conventional method lacks a unified standard.This leads to differences in risk identification results and cannot provide scientific strategies to support grid maintenance decisions.In this regard, it is necessary to overcome the dependence of the conventional identification method on the original data set and use the data flow model as the basis to detect abnormal data by dynamically updating the data in the data flow to achieve effective identification of grid operation risks [4].

Extraction of abnormal features of new energy grid operation data based on linear decision function
Firstly, two new energy grid operation anomalies need to be defined.It is assumed that the new energy grid operation data is circulated in the mode of data stream ( 1, 2,..., ) and the scale of the corresponding sliding window in each transmission line is ( 1,2,... ) , which is a variable that can be adjusted to control the specific scale of the new energy grid operation data in each transmission path.It is assumed that the linear decision function F represents the data characteristics of this sliding window and ( ) i f w represents the normal fluctuation interval in the data stream [5][6].By comparing the linear decision function F with the normal fluctuation interval ( ) i f w , the interval anomaly data can be obtained as shown in the following formula.
( ) The value of the linear decision function F can be adaptively adjusted according to the set threshold.The completed linear decision function needs to be matched with the sliding window scale.When performing data processing of a single power operation data stream, the value of the linear decision function F can be the same as the original data.According to the above formula, the data that exceeds the normal fluctuation interval ( ) i f w can be considered interval anomalous data.In the process of extracting abnormal features of new energy grid operation data, it is usually necessary to judge the abnormal features with the help of the abnormal threshold of data stream fluctuation.Hence, the abnormal threshold of the data stream needs to be selected reasonably.If the threshold range is too large, the abnormal features cannot be extracted.If the threshold range is too small, the extraction error will occur easily.Therefore, this paper combines the main features of the new energy operation data and defines the data stream abnormality threshold as follows.
( ) ( ) ( ) represents the fluctuation interval of the abnormal data of the new energy stream, and ( ) avg x and ( ) std x represent the mean and standard deviation of the sliding window of the data stream, respectively.By adjusting this value, we can control the size of the abnormal threshold of the data stream so that we can effectively determine the abnormal data.When processing simple data streams with low data accuracy, the fluctuation range of the abnormality threshold can be set as an integer so as to improve the efficiency of extracting abnormal data features.
In this paper, the effective extraction of fluctuating abnormal data features is achieved by setting a ratio threshold.For increasing and decreasing data streams, different conditions need to be set to achieve data feature extraction, and the specific formulas are shown.

New energy grid data flow sliding window model construction
Assuming that S represents the new energy grid operation data set with dimension M, the specific formula of this data set is shown. .

.. n S A A
) where X represents the new energy grid operation data stream of dimension m, and each data item in X represents the real-time operation data at the same sampling moment. 1 2  , ,..., n A A A .
The new energy grid operation data set represents different attributes, and each attribute can be clustered as a class.In this paper, the sliding window model is constructed by combining the above data streams, and the structure of the sliding window model is shown in the following figure.1, in the sliding window model, fixed window, data flow, and data flow direction are three important components.Among them, fixed window is a type of window with a fixed window size, such as 1000 pieces of data per window.In the sliding window model, fixed windows are used to divide the data stream into multiple windows and perform data processing within each window.By using a fixed window, infinite data streams can be transformed into limited data blocks for batch processing and real-time analysis; A data stream refers to a continuous data stream composed of a series of data.In the sliding window model, data flow is the basic element of the input model, which contains the raw data that needs to be processed and analyzed.The data stream can come from various sensors, monitoring devices, and controllers, such as output data of solar panels, rotational speed data of wind turbines, etc; The direction of data flow refers to the direction in which data is transmitted in the network.In the sliding window model, data flow is usually bidirectional, ranging from device to central server to device.The direction of data flow plays an important role in the model, as it determines the transmission path and processing method of data, directly affecting the performance and effectiveness of the model.
Through the three components of fixed window, data flow, and data flow direction, the input data flow is divided into a series of data blocks according to the size of the fixed window.Process and analyze each data block and output corresponding results.Sliding window operation is performed between data blocks, where a portion of the data in the previous block overlaps with a portion of the data in the subsequent block to ensure continuity and real-time performance.According to the direction of data flow, the processing results are sent to the corresponding devices or central servers to form a complete sliding window model.

New energy grid operation risk adaptive identification algorithm construction
For the actual characteristics of multiple data stream operation of new energy grid equipment, this paper uses the constructed linear decision function, i.e., the abnormal data extraction function.It is used as the basis to dynamically update the conventional data in the new energy grid operation data through unsupervised learning mode to achieve adaptive identification of operational risk.The flow chart of the adaptive identification algorithm is shown as follows.As shown in Figure 2, the adaptive risk identification algorithm for the operation of new energy grids takes the sliding window data flow from the upper sliding window model as input to generate a summary model.In the model, the active window data is combined with other auxiliary data for initial clustering processing.The data during the initial clustering process is processed and analyzed online to obtain k cluster centers, Input the clustering results, i.e. the clustering centers, into the offline clustering processing model to complete the risk identification output, where k is the number of cluster centers.Hypothesis i M represents the amount of new energy grid operation data under each clustering class.Then, in order to realize the dynamic update of normal data, the expired data in the clustering data need to be eliminated, and the specific processing formula is shown.

Experimental preparation
In order to prove that the proposed linear decision function-based adaptive identification method for new energy grid operation risk is better than the conventional adaptive identification method for new energy grid operation risk, after the theoretical part of the design is completed, an experimental session is constructed to test the actual identification effect of this method.In order to ensure the experimental effect, two conventional new energy grid operation risk adaptive identification methods are selected for comparison, namely the data mining-based new energy grid operation risk adaptive identification method and the data fusion-based new energy grid operation risk adaptive identification method.
The experimental object of this experiment is the historical operation data of the energy grid of a city, and the historical data of the grid in the past three years are retrieved as the original data set for this experiment.By pre-processing the original data set, including data cleaning and data compression, the efficiency of the subsequent algorithm for data processing is improved.The specific parameter configuration of the linear decision function-based risk identification algorithm used in this paper is shown in Table 1.To improve the reliability of the experimental results, five different fault operation data were selected from the original data set for this experiment, which involved high-and low-temperature exothermic faults as well as high-and low-temperature discharge faults.The data were analyzed by using three methods to compare the risk identification effects of different methods.

Analysis of test results
The comparison criterion selected for this experiment is the accuracy of different methods for grid risk identification, and the specific measurement index is the effective data error deletion rate.The lower the value is, the higher the accuracy of the algorithm for risk identification will be.The specific calculation formula is shown.
where k represents the pre-defined number of clustering classes, ' IOP Publishing doi:10.1088/1742-6596/2592/1/0120216 curve of the effective false deletion rate data, it is obvious that the false deletion rate of the linear decision function-based new energy grid operation risk adaptive identification algorithm proposed in this paper is significantly lower than that of the two conventional identification algorithms.The average false deletion rate is about 25%.The average false deletion rate of the two conventional recognition algorithms exceeds 50%, which proves that the recognition accuracy of this paper is better than that of the conventional recognition algorithms.

Conclusion
In this paper, a new adaptive identification method for grid operation risk is proposed to address the problem that conventional risk identification methods cannot accurately identify anomalous data for small fluctuations.By combining the linear decision function, the threshold formula of the abnormal data is constructed to realize the effective extraction of the abnormal data features.The adaptive identification algorithm constructed on this basis has a better identification effect and can effectively identify the bumpy data in the power grid operation data.

Figure 1 .
Figure 1.Sliding window model structureAs shown in Figure1, in the sliding window model, fixed window, data flow, and data flow direction are three important components.Among them, fixed window is a type of window with a fixed window size, such as 1000 pieces of data per window.In the sliding window model, fixed windows are used to divide the data stream into multiple windows and perform data processing within each window.By using a fixed window, infinite data streams can be transformed into limited data blocks for batch processing and real-time analysis; A data stream refers to a continuous data stream composed of a series of data.In the sliding window model, data flow is the basic element of the input model, which contains the raw data that needs to be processed and analyzed.The data stream can come from various sensors, monitoring devices, and controllers, such as output data of solar panels, rotational speed data of wind turbines, etc; The direction of data flow refers to the direction in which data is transmitted in the network.In the sliding window model, data flow is usually bidirectional, ranging from device to central server to device.The direction of data flow plays an important role in the model, as it determines the transmission path and processing method of data, directly affecting the performance and effectiveness of the model.Through the three components of fixed window, data flow, and data flow direction, the input data flow is divided into a series of data blocks according to the size of the fixed window.Process and analyze each data block and output corresponding results.Sliding window operation is performed between data blocks, where a portion of the data in the previous block overlaps with a portion of the data in the subsequent block to ensure continuity and real-time performance.According to the direction of data flow, the processing results are sent to the corresponding devices or central servers to form a complete sliding window model.

Figure 2 .
Figure 2. Flow chart of adaptive identification algorithm for new energy grid operation risk

inFigure 3 .
Figure 3.Comparison results of effective false deletion rateAs shown in Figure3, it can be seen that with the increasing number of algorithm iterations, the effective false deletion rate of different algorithms for data also changes.By observing the change

Table 1 .
Configuration of algorithm parameters in this paper