Research On User Electricity Abnormal Detection Based On Temporal Convolutional Network

With the rapid development of my country’s economic level and the improvement of people’s living standards,the demand for electricity by the people of our country is also rising.Electricity plays a pivotal role in people’s lives and social development.However,along with a large amount of electricity demand comes the increase of transmission and distribution losses in the grid system.Among these losses,in addition to natural line losses,Non-technical losses(NTL)caused by users’ abnormal electricity use behaviors,such as electricity stealing,account for the majority.Electricity bills are an important source of input for national finances.Abnormal user behavior not only affects the country’s economic order,but also theft of electricity by replacing metering equipment as the main means is more likely to cause power system failures and affect the power grid and people’s safety.Therefore,the importance of power abnormality detection has become increasingly prominent.In recent years,the rapid development of smart grids and the popularization of smart meters in China have marked my country’s entry into the era of big data in electric power.Deep learning is an emerging method in the field of machine learning,a powerful tool for processing big data,and has powerful automatic feature extraction capabilities.And the advantages of processing high-dimensional and non-linear data,etc.,have very good research value.This paper proposes a user power anomaly detection model based on deep learning Temporal Convolutional Network,establishes an accurate user power anomaly detection model to improve the accuracy of user power anomaly detection,and provides important for the development of appropriate energy management strategies.The basis for providing advanced technical solutions for the power system in the era of big data.The main research contents of this paper are as follows:First of all,it summarizes and introduces the current abnormal behavior of users’ electricity consumption and the phenomenon of non-technical loss and the research significance,and reviews the research status of domestic and foreign experts and scholars in the field of electricity anomaly detection in recent years.This article introduces the temporal convolutional network model;for the commonly used traditional machine learning methods such as support vector machines,logistic regression,and commonly used deep learning models,such as convolutional neural networks,long and short-term memory networks and other algorithm processes,it has excellent application characteristics.The shortcomings were compared and studied.Secondly,carry out data cleaning,deal with the missing values and outliers in the data set,and carry out the normalization of the data.In addition,in order to solve the problem of the largest power theft data set,that is,the number of abnormal samples in the process of user power theft behavior detection is much smaller than the number of normal samples,the power theft data is a typical example of an unbalanced data set.Through research,this article uses SMOTE(Synthetic Minority Over-sampling Technique)algorithm to process unbalanced data,and introduces the SMOTE algorithm in detail.Finally,introduce and implement the end-to-end user power anomaly detection model based on Temporal Convolutional Network(TCN)proposed in this article under the framework of deep learning.And input the processed data into the temporal convolutional network model for effective end-to-end time series classification.The results show that the method proposed in this paper performs better than the existing support vector machine(SVM),logistic regression(LR),and convolutional neural network(CNN)on the electricity meter data set collected by the State Grid Corporation of China(SGCC).And long short-term memory network(LSTM)and other methods.