Research On Abnormal Electricity Behavior Detection Method Based On ISSA-RF

Some power users use abnormal electricity by various means to achieve the purpose of reducing power charges,which will not only cause security risks,but also destroy the normal market competition order.The electricity information system provides a large number of user energy data,which contains many useful information.How to use the relevant data-driven method to analyze the user energy data and detect abnormal electricity behavior can not only save manpower,but also improve the accuracy of inspection,which has important engineering application value.Firstly,this paper summarizes and analyzes the research status of abnormal power consumption detection by scholars at home and abroad,and analyzes several commonly used detection methods,including Random Forests(RF),Support Vector Machine(SVM),logistic regression model and decision tree model.Based on the measured data set of a power grid,the corresponding parameters of the three models are determined by Sparrow Search Algorithm(SSA),and the effect of abnormal power consumption behavior detection is tested.In the specific test process,due to communication failure and equipment failure,there may be problems such as data missing and data error in meter reading.Therefore,the measurement data obtained from the gateway meter and the user’s electric energy meter are preprocessed first.The statistical characteristic data with strong explanatory power consumption data are extracted.Considering the influence of random parameters on the performance of machine learning model and the small proportion of abnormal power consumption data,SSA is used to optimize the parameters and the area under the ROC curve(AUC)and loss entropy(log-loss)are used as the criteria to judge the quality of the model.The test results showed that the AUC value and loss function of random forests were higher than those of SVM,logistic regression and decision tree models,with AUC value of 0.9265 and log_loss value of 0.197.Therefore,the random forest model is selected as the main research object,and the super-parameter optimization method is studied in depth.Aiming at the shortcomings of SSA,such as easy to fall into local optimal iteration and single convergence in the later stage,an abnormal power consumption behavior detection method based on improved SSA search algorithm and random forest is proposed.This method takes the top 10%of the sparrows as the elite sparrows at the end of the sparrow search,compares the sparrows after the search with the elite sparrows,and takes the elite sparrow as the fitness value.Furthermore,the application of improved SSA algorithm and original SSA algorithm in random forest is simulated and compared.The results show that the accuracy of ISSA-RF is 5.2%higher than that of SSA-RF,and is higher than that of RF,SVM,logistic regression and decision tree models.It can be seen that the improved SSA algorithm combined with random forest can effectively improve the accuracy of abnormal power consumption detection by expanding the search area of the algorithm and reverse search.