Research On Non-invasive Load Feature Extraction And Identification Algorithm

Facing the technical demand of intelligent power utilization for meticulous load data and electricity service behavior mining,a non-intrusive load identification method based on co-evolutionary multi-objective particle swarm algorithm was proposed to obtain better load identification effect.Firstly,the basic framework of the non-intrusive load monitoring(NILM)system is introduced.The extraction methods of steady and transient characteristics of appliances are analyzed.In order to extract more useful information from the limited training data,the mutual information method is used to select the most effective load features related to the appliance identification.Five electrical features are selected to form a multi-feature input space.Secondly,the load decomposition methods are introduced from mathematical optimization and pattern recognition.The applicability of K-nearest neighbor algorithm,affinity propagation clustering algorithm and genetic optimization algorithm in NILM is studied through experiments.Aiming at solving the problem of insufficient accuracy with single feature optimization,a NILM method based on co-evolution multi-objective particle swarm optimization(CMOCPSO)algorithm is proposed.The sub-populations are used to find the local and global optimal solutions of each sub-population,and then the multi-objective optimization and identification are performed.Using public data sets,the proposed algorithm is verified experimentally.Compared with other methods,the result proves the algorithm is effective.Finally,to solve the problem that loads with similar electrical characteristics are prone to misidentification,a method for correcting based on the load switching probability distribution curve is presented.The method uses back propagation neural network to obtain the switching probability distribution curve of each load,and then modifies the recognition result according to the pattern of load operation.Facing the technical demand of intelligent power utilization for meticulous load data and electricity service behavior mining,a non-intrusive load identification method based on co-evolutionary multi-objective particle swarm algorithm was proposed to obtain better load identification effect.