Study Of Event Detection Methods For Non-intrusive Load Monitoring

For reducing the shortage of energy supply and the development of smart grid,the strengthening of power consumption monitoring in the house has important practical significance.In order to improve the utilization of energy in residential environment,visual monitoring of household energy consumption is an important technology.In residential buildings,electricity monitoring helps to save energy,and non-intrusive load monitoring(NILM)technology is an important method for the realization of electricity monitoring.The purpose of NILM is to decompose the total power consumption into single devices’ power consumption,using machine learning techniques.In this thesis,the event detection methods for non-intrusive load monitoring(NILM)are studied.The so-called events refers to ON/OFF operation of appliances.Event detection plays an important role in non-intrusive load monitoring to accurately detect the switching of appliances in a residential environment.Improving the detection ratios and keeping the robustness of those methods while keeping the computational cost under control is important.The power signal for BLUED data has noise and has a great influence on the result of the event detection.Firstly,In this thesis,median filter is used to preprocess the total power data,and the best total power data is obtained.Secondly,the existing standard χ2 GOF event detection method is introduced,which is not robust to differences in base load power consumption.Aiming at the problems of standard GOF method,two novel and robust methods are studied:a modified version of the chi-squared goodness-of-fit(χ2GOF)test and an event detection method based on cepstrum smoothing.Then,through the MATLAB simulation software,BLUED data sets are used to simulate the two methods respectively and their simulation results are analyzed.Finally,according to the analysis results,we use the characteristic measure F to analyze the robustness of the three event detection methods respectively,and draw the conclusion that the two new event detection methods are robust.Therefore,the example simulation results based on the BLUED dataset show that the two research algorithms outperform the standard GOF method for higher base load.