Research On Key Technologies Of Load Decomposition Of Household Appliance

Non-intrusive load decomposition technology is the key research content of intelligent power consumption system,which can deeply analyze users’ power consumption information under the background of ensuring users’ privacy.In the load decomposition,the main problems are low power consumption,similar power appliances can not be identified.At first,in order to solve the above problems,researchers introduced high-frequency features for recognition,but the corresponding calculation is complex,and the actual implementation is more difficult.With the development of pattern recognition in recent years,the method of pattern recognition is widely used in non-intrusive load decomposition,but researchers ignored the usability of the method.To solve the above problems,this paper focuses on the application of combination optimization method in non-intrusive load decomposition.(1)This paper sorts out and classifies the existing research methods of nilm from two aspects of load characteristics and load identification algorithm,discusses the advantages and disadvantages of combination optimization method and pattern recognition method respectively,and determines the evaluation index of nilm algorithm,which lays the foundation for the construction and evaluation of non-intrusive load decomposition algorithm proposed in this paper.(2)High frequency features are introduced into sparse classification,and a non-intrusive hierarchical load decomposition method based on steady-state features is proposed.Firstly,the mean shift algorithm is used to cluster the electrical data,and the number of working states,power and current harmonics of each electrical device are obtained.The feature dictionary is constructed based on the active power,reactive power and current harmonics,and the sparse matrix is constructed based on the electrical state.Secondly,in order to improve the accuracy of load decomposition and reduce the computational complexity,the idea of hierarchical classification is used for load decomposition.Firstly,the active power is used for identification matching,then the reactive power is used for identification matching,and finally the load decomposition result is calculated based on the Euclidean distance and the high frequency current harmonics of electrical appliances.Experimental results show that this method improves the accuracy of load decomposition compared with the traditional sparse decomposition method.(3)In the non-intrusive load decomposition,a hierarchical load decomposition is performed by combining transient with steady-state features.In the training process,clustering analysis is performed on the current data to construct a feature dictionary and rarefaction matrix of steady state versus transient,and feature weights are calculated from the mutual information of current harmonics.In the test process,the transient features were extracted in the way of secondary windowing treatment of the current data.The sparse matrix is dimensionally reduced according to the transient features and the current state at the previous moment.After the active power identification and the reactive power identification using the sparse matrix after dimension reduction,the high-frequency current harmonic identification was performed according to the calculated characteristic weights of the mutual information of the current harmonics to obtain the load decomposition results.Experimental results show that this method improves the accuracy of load decomposition compared to the load decomposition of steady-state features.