Non-Intrusive Load Monitoring Based On Machine Learning

Non-Intrusive Load Monitoring(NILM)technology is an essential direction in developing intelligent grids,enabling the monitoring of individual load operating states,energy consumption,and other information without altering the customer’s circuit structure and providing technical support for information interaction between power companies and customers.Nowadays,load types are becoming more complex and diversified,and existing research primarily uses new algorithms and introduces multiple load features to improve the monitoring effect.In contrast,redundant features can affect monitoring efficiency.In this thesis,we focus on load feature selection and load identification algorithms,and the specific work is as follows:(1)The advantages of non-intrusive load monitoring and intrusive load monitoring are compared and analyzed,the basic process of non-intrusive load monitoring is described,the technology of the detection process including data collection pre-processing and machine learning algorithms for monitoring is explained,the event detection technology used in this thesis is determined,and the evaluation system of load monitoring results is constructed.(2)The principle of load identification is analyzed,the electrical characteristics used for load monitoring are summarised,a method for selecting load characteristics based on the K-nearest neighbor algorithm is proposed,the correlation of load characteristics is calculated by using cross-validation,several typical load operating characteristics are analyzed by means of arithmetic examples,and the selection of load characteristics is completed.(3)A non-intrusive load identification method based on Boostrap aggregating(Bagging)ensemble learning is proposed to improve the identification effect of the model through the selection of a base classifier,the construction of an integration strategy,and the optimization of base classifier parameters.Several sets of numerical experiments are designed to verify the effectiveness of the proposed algorithm for load monitoring and to analyze the impact of load event interval on the identification method used in this thesis.The application scenarios of load monitoring are analyzed in terms of the itemized electricity bill calculation and abnormal power consumption of users,which provide references for users to optimize power consumption,safe power consumption,and the grid to improve the incentive power consumption mechanism.