The Research On Non-Intrusive Residential Load Monitoring Based On Event Detection

Non-intrusive Load Monitoring(NILM)refer to collecting the voltage and current waveform data of electrical energy consumers,by the data processing and feature analysis to complete the nature of the user’s electrical equipment classification and the identification of major electrical equipment.In terms of power consumers,the NILM can make the power consumption of various types of electrical equipment more detailed,which guides the users to actively participate in the "peak-to-peak" of the grid within the mechanism of price incentive.As for the grid,it is more realistic to understand that the composition of the power system,and providing basis for improving the efficiency of electrical energy.So it has great academic and practical significance.In this thesis,the working voltage and current waveform data of the typical residential electricity loads are collected.Based on the analysis of the electrical characteristics of the electricity load,the two-side cumulative sum based on sliding window and the C4.5 decision tree method are used to carry out the electricity load switching event detection,the nature of electrical equipment classification and the main electrical equipment identification.the main work include:The working voltage and current waveform data of the typical residential electricity load were collected and the steady-state electrical characteristics of the electricity load are studied by Fourier harmonic analysis.This thesis divides the experimental household electrical loads into resistive class,reactive class,complex class and electronic class,and evaluates the feasibility of NILM based on steady-state electrical characteristic parameters.The two-side cumulative sum based on sliding window method is used to detects electricity load switching event.The switching event refers to the electrical behavior of the electrical load inputted or removed.Due to features including the wide range of active power values of typical residential electricity load,the duration of the transient process and different from of the power verious,and the active power fluctuation being large in some electrical equipment work,the active power of the fundamental wave is filtered,plus an improved algorithm for two-side cumulative sum based on sliding window method is proposed to improve the electrical adaptability of event detection.In order to make good use of the advantages of different feature parameters,the C4.5 decision tree algorithm for interclass identification and the k-nearest neighbor learning algorithm for within class identification are employed.Based on the C4.5 decision tree method,the data of the fundamental power factor,Subtriction of the third harmonic ratio of current-voltage and the current second harmonic ratio of 24 experimental objects are used as the training data set,and the characteristic parameters based on the information gain rate,the order of the internal nodes of the decision tree and the threshold of the splitting attribute are determined,and the model of load decision tree class recognition is established.Based on the Euclidean distance,the k-nearest neighbor learning algorithm is used to calculate the fundamental active power reactive power phase plane within the load class recognition.The results indicated that the method is feasible and propagable by using experimental data,home field test data and international BLUED data set.