Research On Non-intrusive Load Decomposition Based On Big Data

Smart grid and energy conservation are important mehods and major requirements for the intellectualization of power grid in China.The implement of real-time monitoring in power system is an important and mark of smart grid.Therefore,Non-intrusive load decomposition technology emerges as the times require.Nowadays we are in the age of“Big Data”,so people from all walks are searching for means of analysis and decision-making via Big Data technology,and especially with the continued perfection of power system and data acquisition system in our country,the annually cumulative power data in our country will reach PB(2⁵⁰bytes)grade.Thus it is inevitable to apply big data technology to the study of non-intrusive load decomposition.This thesis presents that Big Data technology can be used in the analysis of non-intrusive load decomposition issue and the main points are listed as follows:?1?Digital signal processing.In the process of data acquisition,the acquired signals mixed with noises due to various reasons will influence analysis results.Electrical signals are denoised by the method of wavelet denoising and acquired the best de-noising effect by choosing the optimum parameters based on a number of tests.?2?Non-intrusive load decomposition in the steady-state process.The steady-state process in this thesis is defined as non-load switching in the power system.The decomposition process is divided into two parts,training and actual measurement,through non-event detection method.By decision tree algorithm,what kinds of load in the single load operation is accurately identified.The least squares algorithm on the basis of load linear constraints is presented and also the amount of each load in the multi-load combined operations is precisely calculated.?3?Non-intrusive load decomposition in the transient process.In this thesis,the transient process is defined as a transitional period from one steady-state process to another caused by load switching in power system.Firstly,the decomposition process is divided into two parts according to the event-based detection method.Then,by choosing 13 parameters as the load signature to characterize different loads,a PCA dimensional reduction method based on Min-Max standardization is induced to reduce the dimensionality of High-dimensional feature space.In accordance with K-means clustering algorithm,the load is divided into different types,and the identification accuracy of different load transient processes on the basis of decision tree algorithm reached 87.5%.?4?Further study on non-intrusive load decomposition in the transient process.To improve the identification accuracy of different load transient processes,Random Forest algorithm,as a strong classifier,is used in the load transient process identification,and K-means improved algorithm on the basis of maximum distance and initial clustering center is presented to avoid getting into local optimum led by initial clustering center randomization,so the final identification accuracy in the transient process can reach 100%.This thesis employs the data collected in the tests to verify the method of non-invasive load decomposition in the steady-state process and BLUED data set publicized by Carnegie Mellon University to verify the method of non-intrusive load decomposition in the transient process,both of which get the good effects.A summary on the work of this thesis and an outlook for the future study are presented at the end of this thesis.