Analysis Of Abnormal Electricity Consumption Behavior Of Users Based On Neural Network

At present,China's smart grid construction of rapid development,information acquisition system is gradually perfect,can have the power of the collected data,the user of electricity in the data information,through analyzing the power use data effectively,reasonably mining the useful information in electricity data,access to relevant laws and patterns of different kinds of power information,provide the basis for electric power dispatching and power services.At the same time,with the development and expansion of the power grid,abnormal electricity consumption behavior is increasing day by day.Abnormal electricity use not only brings huge losses to power grid companies,but also damages to power grid facilities and causes huge hidden dangers to the safe operation of power grid.It can be seen that abnormal electrical behavior detection is an important link to ensure the safe operation of power grid.Through the study of the above problems,the main means of abnormal power consumption are power metering device fault,power theft,power system harmonic interference.This article,from the perspective of the metering device fault and power two big reasons,further to explore the characteristics of fault with the power metering device,from electricity metering device test data and user data,the characteristics of the data to sort and calculation,and then USES a limit of machine learning algorithm based on principal component analysis to improve,the analysis of the metering device test data,to classify all kinds of metering device failure.Then,particle swarm optimization(PSO)was used to optimize BP neural network algorithm,and the known abnormal data was used to train the algorithm,and the abnormal electricity consumption model was built to identify the abnormal electricity consumption behavior of users.The main research contents are as follows:(1)through the analysis of the principle of the electric energy metering system,analyzed all kinds of electric energy metering device,the characteristics of the fault and through to the electric power metering device equipment detection circuit of data collection,using a limit of machine learning algorithm based on principal component analysis to improve,the analysis of the metering device test data,for all kinds of metering device fault classification,identify the faulty user electric energy metering device.(2)This paper focuses on nine aspects including peak active power,flat active power,valley active power,total reactive power,total active power,power factor,threephase current imbalance rate,three-phase voltage imbalance rate and three-phase voltage offset to evaluate and predict the suspected power theft coefficient of target users.By using BP neural network algorithm,the user's historical electricity characteristic data as input,the actual cases of power theft results(power is 1,otherwise 0)as the target output,the data into the BP neural network for repeated training,after training,the test data set into the trained BP neural network,the judgement of the resulting output power value.In the experimental stage,through the collection and analysis of data of 300 household metering devices and electricity consumption data of users in a certain area of Yunnan Province,90% of the data was trained and modeled,and 10% of the data was selected for testing.The results show that in the verification of the PCA-ELM algorithm for the fault detection of metering devices,the fault detection accuracy of metering devices reaches 100% with an average error of 4.5%.At the same time,in the verification of PSO-BP algorithm for users with abnormal power consumption,the network verification accuracy reaches 100%,and the simulation test results are consistent with the actual abnormal power consumption,with an average error of 2.87%.Finally,through the cross comparison of the two conclusions,the users with metering faults are excluded from the suspected abnormal electricity users,and the complete list of power theft users is finally determined.