Research On Non-Intrusive Load Monitoring Method Based On Load State

With the deepening of research on smart grid technology in China,smart meter,as the foundation of Advanced Measurement Infrastructure(AMI),has been extended to the end of the grid.Smart electricity meters can realize the collection and analysis of users’ electricity data for a long time.At the same time,combined with 5G communication technology,they can realize large-scale and massive terminal parallel electricity data communication.Excavating the details of electricity consumption contained in electricity consumption data and studying the type and composition of users can help electric power enterprises to provide personalized services for consumers,improve fault detection and formulate energy incentive measures.In addition,these details can encourage energy-saving behavior.Therefore,electricity load monitoring is of great significance to the formation of smart grid technology,the improvement of energy utilization rate and the construction of energy saving society.Non-intrusive Load Monitoring(NILM)achieves this goal by simply installing smart meters at the power entrance and analyzing the total power signals collected to obtain detailed operational information about individual loads in the system.The main research contents of this thesis are as follows:(1)In order to obtain the time point of load state switching,an event detection algorithm based on sliding window Pettitt detection was proposed to determine the load switching point.In addition,a clustering algorithm suitable for load state extraction is improved to obtain the total number of states of finite multi-state loads and the clustering center of each state,which lays a good foundation for load identification algorithm.(2)Aiming at the identification of finite multi-state load,according to the principle that each running state of electricity loads has its unique load imprints,this thesis firstly obtains all load states by using clustering algorithm,and then extracts the corresponding imprints to form load feature library.An individual coding method based on multi-state was proposed to optimize the objective function.Finally,NSGA-Ⅱ,a fast non-dominated genetic algorithm,was used to solve the model.(3)Aiming at the identification of continuous variable state load,this thesis extracts the transient active power of load switching,calculates the peak value of power impact,duration,power difference before and after impact,and the power curve integral between the start point and the end point of impact as load characteristics,and then uses XGBoost algorithm to identify the load.The analysis of numerical examples shows that the proposed methods have higher recognition accuracy.