Research On Non-intrusive Residential Load Monitoring Metho

With the proposal of the "3060" dual-carbon goal and the advancement of demand-side digital construction,the demand for monitoring load and electricity usage information has become increasingly prominent.NILM can identify residential electricity equipment and their status,enabling users to understand their electricity usage information and guide them in formulating reasonable electricity usage strategies.At the same time,it helps with fine management and optimization of the demand side of the power grid,achieving the goal of saving electricity.Its research has important significance for demand-side energy conservation and emission reduction and digital level improvement.To address the shortcomings of the existing non-invasive load decomposition methods,such as incomplete and comprehensive decomposition of independent loads leading to incomplete electrical information,and the problem of low recognition accuracy of load features under low-frequency sampling resulting from feature overlap,research was conducted on load decomposition and load identification algorithms.The main work is as follows:(1)A load decomposition method based on VMD-FastICA was proposed.Considering the advantages of VMD in signal analysis and processing,an underdetermined blind source separation method based on VMD was adopted to process mixed data collected by non-invasive load monitoring based on the power of load.The data of each load running separately was separated.Firstly,the mixed signal power is decomposed into several IMF through VMD algorithm,and then,the signal is expanded by kurtosis criterion and singular value decomposition.A virtual channel signal is introduced to construct a new mixed signal as the input matrix of FastICA algorithm,thus transforming the underdetermined blind signal into overdetermined blind signal separation.Finally,the source signal is separated by using the FastICA algorithm.Compared with EMD and EEMD algorithms,the proposed algorithm was verified to have better decomposition performance.(2)On the premise of load decomposition,a load identification method based on VMD-EntropyLSTM was proposed.Taking advantage of the VMD energy entropy in the field of feature extraction and the strong ability of LSTM network in processing time series,VMD energy entropy was used for load feature extraction,and LSTM was used for load identification.Firstly,independent load signals were decomposed by VMD into various IMF from high to low frequency,and the energy and energy entropy features of each IMF were extracted to construct a multimodal feature matrix.Then,a load feature library was established to construct an LSTM classifier.Next,VMD energy entropy was extracted from the load signal decomposed by VMD-FastICA,and the result was input into the LSTM model to achieve load identification.Finally,the performance indicators of load identification were compared and verified with other classification algorithms such as GRU,CNN,and SVM,showing that using LSTM had better performance in load identification.