Research On Non-Intrusive Family Load Monitoring Method

With the increasing prominence of energy problems around the world,energy saving and consumption reduction has become a common problem for both power grid and users.Non-intrusive load monitoring technology(NILM)puts forward a new solution to save energy and reduce consumption.By monitoring the data of the internal electrical load of users,it realizes the energy analysis of users,so as to provide power supply strategy and electricity price strategy guidance for the power grid,provide energy analysis for users,improve users’ electricity consumption behavior,and achieve the purpose of saving energy and reducing consumption.In addition,NILM can still provide guidance and suggestions for some special scenarios in social life,such as electrical fire detection,monitoring of sensitive users,abnormal electricity behavior.This paper puts forward two solutions based on event and not based on event for the research of non-intrusive load monitoring.the main research contents are as follows:1)Build a load data acquisition platform to collect the voltage and current data of electrical appliances.Using the measured data,the feasibility and effectiveness of sliding window bilateral cusum algorithm as an event detection algorithm are studied,which lays a foundation for subsequent load monitoring.2)In view of the low accuracy of load identification due to load feature similarity and unreasonable feature selection,a new load feature extraction method is proposed,which extracts 29-dimensional time-frequency signal features to lay the foundation for load identification,and Relief F algorithm is used to optimize the extracted time-frequency features to reduce the influence of irrelevant or poor classification performance features on load identification.Genetic algorithm(GA)is used to optimize the weight and bias of extreme learning machine(ELM)to improve the performance of extreme learning machine.The experimental results show that GA-ELM can effectively improve the accuracy of load identification than ELM,and the proposed 29-dimensional time-frequency features are optimized by Relief F features to significantly improve the accuracy of load recognition.3)Aiming at the problem that the high data sampling frequency leads to the high pressure of the data acquisition terminal,the load identification of the measured data after the sampling rate reduction is studied,and the effectiveness and applicability of the non-intrusive load identification method based on Relief F feature optimization and GA-ELM after the sampling rate reduction is verified.On this basis,the event detection algorithm and load identification algorithm are combined to experiment on the measured data,and the event-based non-intrusive load monitoring is realized.4)Aiming at the problem of poor load decomposition accuracy caused by not considering the running state of electrical appliances in the objective function design of non-intrusive load monitoring,in this paper,a new objective function based on active power,reactive power,sparse change of running state and running probability of electrical apparatus is proposed to decompose the load.The experimental results verify the improvement of the decomposition accuracy brought by the proposed objective function.Finally,the improved differential evolution algorithm(j DE)is applied to load decomposition,and the proposed new objective function is used to decompose the load.The experimental results show that the proposed j DE algorithm further improves the accuracy of load decomposition and realizes non-intrusive load monitoring without event.