Research On Technologies Of Electricity Stealing Behavior Laboratory Reproduction And Anti-electricity Stealing Intelligent Judgment

The behavior of stealing electricity seriously damages the economic interests of electric power enterprises,which poses a great threat to the market order of electricity consumption,the stable operation of power grid and the safety of people’s personal property.With the continuous improvement of the degree of intelligence and informatization of power system,the power companies can obtain exponentially increasing amounts of data which providing strong support for data-driven methods.The use of intelligent algorithms to identify electricity stealing users can improve the accuracy of electricity stealing inspections and reduce the blindness of detection.Further identify the power stealing methods of suspected power stealing users,make the on-site inspection process of power stealing accurate and efficient,while establishing the image of the power grid and deterring users of electricity theft.Therefore,the research on the detection method of electricity stealing with high accuracy and the discrimination method of electricity stealing behavior with strong guidance can effectively restore the economic loss of power enterprises and improve the security and reliability of the power grid.Firstly,the thesis discusses the serious influence of the behavior of stealing electricity on electric power enterprises,market electricity consumption,power network security and people’s personal and property safety,expounds the importance and necessity of the research on the detection of stealing electricity and the identification of the behavior of stealing electricity,introduces the research status of the detection of stealing electricity and the identification of the behavior of stealing electricity in detail,analyzes and summarizes the advantages and disadvantages of the existing research methods.To solve the problem of lack of data sources for current electricity theft analysis.research on the multi-category laboratory reproduction technologies of electricity stealing behavior.This thesis builds a multi-category laboratory reproduction platform for electricity theft according to the theory of electricity theft and the measure of electricity theft.For single-phase and three-phase users,reproduce electricity stealing methods such as undervoltage method,undercurrent method,phase shift method,virtual load is used to simulate the user’s electricity consumption behavior,implement electricity stealing by control solid state relays,construct normal electricity branch and electricity stealing branch to simulate normal and electricity stealing users.Built the system software with functions of controlling electricity stealing reproduction platform and electricity stealing behavior discrimination,the software control the electricity stealing behavior laboratory reproduction platform to realize the operation of stealing electricity and obtain electricity stealing data,obtain the discrimination results of the electricity stealing behavior discrimination model and display them in real time.realize the software and hardware design of electricity stealing generation and behavior analysis.In allusion to applicability of unsupervised methods in context of user-side electricity theft detection,this thesis studies how to solve the decoupling problem between feature extraction and anomaly detection,and proposes an electricity theft detection method based on Deep Auto encoder Gaussian Mixture Model(DAGMM).Firstly,the electricity consumption data dimension with stationarity is obtained according to Augmented Dickey Fuller test,Then,the potential features that characterize the user’s electricity consumption habits and the difference with other users’ electricity consumption habits are extracted through the compression network,estimation network and gaussian mixture model are used to obtain sample energy which reflect the degree of anomaly,network parameters are optimized jointly based on end-to-end learning to avoid model decoupling,identify the sample energy exceeding the abnormal threshold set by boxplot as abnormal,accordingly,stealing electricity of users can be detected.The experimental results show that the proposed method is less affected by proportion of electricity theft,the extracted features can effectively reflect the electricity consumption pattern of users.Compared with the existing methods,the accuracy,precision,true positive rate and false positive rate of the proposed method are significantly improved.Aiming at the problem of identify method for stealing electricity behavior guidance can’t meet the current demand,based on the data of electricity stealing behavior multi-category laboratory reproduction platform,we proposes a method for identifying electricity stealing behavior based on the Classification And Regression Tree(CART),extracting voltage amplitude,current symbol,power factor and meter type as input features,using CART to classify the stealing behavior,solve the problem of model overfitting by pre-pruning combined with cost complexity pruning,the category with the most judgments among the accumulated multiple judgment results is used as the final electricity stealing behavior judgment result and output the judgment probability.so as to realize the discrimination of electric theft behavior.The experimental results show that the model proposed in this thesis can capture the interaction between variables and effectively distinguish different electric theft behaviors.Finally,it summarizes the main research work of the thesis,and points out the shortcomings of the research and the improvement direction of follow-up research.