Research On Identification Of Electricity Theft Way Based On The Data Of Electric Energy Data Acquisition System

Although electricity theft identification methods such as big data,unsupervised learning,and neural networks can efficiently identify whether electricity is stolen or not,due to the limitations of data quality and the single type of data used,they cannot provide critical information on which way of the electricity theft is.The purpose of this research is to break the data limit,generate electricity theft data with reliable labels for different ways of electricity theft,build a deep neural network model that can identify different ways of electricity theft,and provide reliable targeted information for electricity theft investigation and punishment to ensure accurately,quickly and effectively deal with electricity theft,and improve the work efficiency of anti-electricity theft.With the powerful circuit and power system simulation function of Matlab,a power theft simulation model is built.Using the GUI and built-in functions provided by Matlab,the parameter setting program of the electricity theft simulation model is designed,and using the static load to analyze the correctness of the electricity theft simulation model.The batch data generation program is written based on the electricity theft simulation parameter setting program.The data normalization is performed in two ways: single-column data normalization and the same type of data combination normalization.Under the Baidu Paddle Paddle deep learning framework,the Sigmoid and ReLU activation functions are used to build a deep neural network model to identify electricity theft ways.The optimal model is obtained by comparative analysis,and the test data set is used to evaluate the accuracy of the optimal model for different ways of electricity theft data identification.The modules in the electricity theft simulation model,such as short connection of current circuit,reverse of current circuit,loss a phase voltage of voltage circuit,misphase of voltage circuit,transformer saturation and transformer ratio changed are completed,and the parameter setting program of electricity theft simulation model is completed,too.Under static load,the waveforms of various electricity theft ways are consistent with the analysis results.The batch generation program of electricity theft data is programed,171060 training sets are generated and 8830 test sets are generated.According to comparing the results of the electricity theft way neural network identification model,the evaluation accuracy of the test data set,using the model of the same type of data combination normalization and ReLU activation function,is the best when the epoch is 49,and the identification accuracy of the model is 0.894 under the optimal parameters.The electricity theft simulation model can correctly simulate the different ways of electricity theft.Such as current circuit short connection,current circuit reverse,voltage circuit loss a phase voltage,voltage circuit misphase,transformer saturation and transformer ratio changed.The electricity theft simulation parameter setting program can quickly and accurately set various electricity theft parameters of the electricity theft simulation model.Based on the data of Electric Energy Data Acquisition System and dynamic load module,the electricity theft batch data generation program can generate a large number of electricity theft data when preset parameter values in advance.The generated electrical data and corresponding label data are accurate,which ensures the data quality and saves a lot of time.Under the same normalization method,the model using ReLU activation function is more accurate,and under the model using the same activation function,the model is more accurate when using the same type of data combination normalization method.There are optimal parameters in the model under different normalization methods and different activation functions.Under the optimal parameter model,the model can accurately identify different electricity theft ways and give the probability values of different electricity theft ways,which provides a critical guidance information for anti-electricity theft work.