Research On Data-driven Residential Electricity Consumption Behavior Characteristics Mining Method

In the context of the smart grid,the widespread application of smart meters and electricity consumption information collection systems has resulted in massive amounts of electricity consumption information on the electricity consumption side.Resident users as an important object of power services,based on these data,in-depth mining of power consumption behavior characteristics can help power companies perceive users’ power consumption habits and is important for carrying out load forecasting,demand response,and electricity price decision-making.Therefore,this thesis focuses on the mining of the characteristics of residential users’ electricity consumption behavior,including three aspects: load curve classification,correlation analysis and electricity theft detection.The main work is as follows:1)Considering the characteristics of large scale,low value density and class imbalance phenomenon of user load data,an algorithm combing improved K-means with long short-term memory network(LSTM)and convolutional neural network(CNN)classification model is proposed.Firstly,an auto-encoder based on LSTM(LSTM-AE)is used to extract load characteristics from high dimensional data.Secondly,to improve the classification accuracy of the K-means on imbalanced data,a method of relative k-nearest neighbor density peaks(RKDP)is proposed to select the initial clustering center of K-means.Finally,a classification model based on the LSTM and CNN networks is used to realize the classification of large-scale load profiles.Experiments have been designed to verify the effectiveness of the proposed method by using public datasets.2)A correlation analysis method of electricity consumption behavior based on association rule mining algorithm is proposed.First,multiple electricity consumption features are extracted from load data covering electricity consumption,electricity consumption stability,potential of peak load shifting and typical electricity consumption patterns.Then,association rules between electricity consumption features and household characteristics are obtained by FP-Growth algorithm to explore the relationship between household characteristics and users’ electricity consumption behavior.Through the correlation analysis of users’ power consumption behaviors affected by multiple factors,mining strong related factors can help power companies understand the characteristics of different types of users’ energy consumption,and have important guiding significance for the development of load forecasting and demand response.3)In terms of electricity theft detection,traditional feature extraction methods are difficult to effectively distinguish the differences between honest and dishonest users,and there is a problem of class imbalance in detection.Therefore,a detection model based on feature fusion and Balanced-bootstrapping strategy is proposed.In the aspect of feature construction,the numerical features and the power consumption pattern features are extracted from load data as the input feature of the proposed model.In the aspect of model design,by using the decision tree model,the ensemble model is constructed through the Balanced-bootstrapping strategy based on random under sampling.The proposed model is validated with real-world smart-meter data.The results show that the model can realize high-precision detection of class unbalanced electricity theft datasets.