Research On Non-intrusive Load Decomposition Method

User load monitoring is a key technology for realizing two-way interaction of smart grids and is receiving increasing attention.Non-intrusive load decomposition analyzes the user’s electrical information and decomposes and obtains the working status of each user’s electrical equipment.It has become the preferred technology for user load,especially for residential user load monitoring.The research of non-intrusive load decomposition can not only enable users to better understand their own power consumption information,but also provide data support for demand response of the grid,creating conditions for the grid to improve scheduling flexibility and enhance renewable energy consumption.Therefore,the research of non-intrusive load decomposition has obvious theoretical significance and engineering application value.In this paper,the steady-state characteristics and transient characteristics of residential electricity load are analyzed respectively.The advantages and disadvantages of the two characteristics applied to non-intrusive load decomposition are summarized.An improved method is proposed for the existing identification methods.A non-intrusive load-decomposition decision tree algorithm that combines steady-state features and transient features.The main work includes:Firstly,for the problem that the non-intrusive load decomposition algorithm based on steady-state features is difficult to identify the unknown load,an improved algorithm combining template matching and KNN neighbor algorithm is proposed.By finding the closest load composition,the untrained load is reduced.The effect is to improve the accuracy of the load decomposition results.The current data of typical household appliances load is used to obtain the steady-state characteristics by Fourier analysis,and the proposed algorithm is verified.Secondly,for the problem that the dimension of the transient feature data with high information is not easy to calculate,after extracting the typical transient feature to form the transient feature vector space,the principal component analysis method is used to reduce the data and try to ensure the original data characteristics.With good degree of reduction,the load is finally classified by KNN neighbor algorithm and evaluated by loss error analysis.The transient processes of different states and kinds are collected in the example verification,and the algorithm is proved to be effective for non-intrusive load decomposition.Finally,in order to further utilize the characteristics of different features and improve the efficiency of non-intrusive load decomposition algorithm,a non-intrusive load decomposition decision tree algorithm combining steady state features and transient features is proposed.The decision tree algorithm combines the advantages of the two features,and adds wavelet analysis in the classification process to further refine the load of difficult classification,and improve the accuracy of load decomposition.The load decomposition simulation is carried out by the experimental data collected,which proves the feasibility of the method.