Research On Power Behavior Analysis Method Based On Load Data Preprocessing

User behavior analysis is very important for user behavior planning and resource collaborative optimization.Existing user behavior analysis methods have the disadvantages of strong subjectivity and physical model that are difficult to deal with the randomness and uncertainty of user behavior in complex grid environment.At the same time,bad load data may appear in the process of distribution network load data collection,which will affect the accurate analysis of user behavior.Electricity theft users in the smart grid will cause huge economic losses every year,hindering the normal operation of distribution network and the security of the power system.In order to solve the above problems,this paper proposes an analysis method of electricity consumption behavior based on load data preprocessing.The simulation results on the real load data sets show that the proposed algorithm has superior performance compared with the existing algorithms.The main research work of this paper includes three aspects.(1)In order to accurately identify and correct the bad data in the process of load collection,so as to improve the reliability of user power consumption behavior analysis,this paper firstly analyzes the power consumption characteristics of different users,introduces the causes and types of bad data,and puts forward a bad data identification and correction method based on combined statistical model and curve similarity.Firstly,the bad data is identified by setting the normal load threshold and the load change rate judgment criterion.On this basis,the gray correlation coefficient is used as the similarity value,and the bad data is replaced by the bad data correction method based on the curve similarity,which improves the availability of the data set.The simulation results on two real power consumption data sets show that,compared with the existing algorithms,the bad data identification and correction method proposed in this paper can accurately identify more types of bad data,and the corrected data obtained by the model is closer to the real data.(2)In order to better analyze user behavior in complex power grids,this paper first proposes an adaptive feature selection algorithm of user behavior based on stacked autoencoder and unsupervised learning to select typical user behavior features and reduce the complexity of user behavior data.Then a user behavior analysis model based on adaptive feature selection and improved clustering is established,and the improved K-Means clustering algorithm is used to realize the unsupervised learning of user behavior data.The experimental results on two real electricity consumption datasets and a public EV charging dataset show that the proposed algorithm has higher feature selection rate and better unsupervised classification performance.The algorithm analysis conclusion provides a theoretical basis for user planning and marketing in the future.(3)In order to improve the detection rate of abnormal power users in smart grid and ensure the safety of power grid operation,this paper proposes an abnormal power consumption pattern detection algorithm based on HRF-WSVDD.Firstly,the random forest algorithm is used to reduce the dimensionality of the user’s daily load data and statistical features,and the features that can best represent the user’s electricity consumption behavior are extracted.Then the abnormal electricity consumption pattern detection model is constructed,use the feature importance value of the feature as the weight,and build a feature weighted support vector description classifier to realize the effective identification of abnormal power users.The simulation results on the EUNITE dataset show that,compared with the existing algorithms,the abnormal power consumption pattern detection algorithm proposed in this paper has better feature representation ability and higher detection rate of abnormal power users,which can ensure the operation of the power grid safety.