Research On Residential Power Consumption Behavior Based On Typical Load Pattern

With the aggravation of energy consumption and the continuous increase of power demand,the power system is becoming more and more complex,and the power supply is becoming more and more complex and diversified.The study of residential power load mode can promote good power supply behavior,reduce economic costs,and improve power supply efficiency.The power supply platform can supply power in stages according to the user’s load mode,which has guiding significance for the follow-up of TOU price and the establishment of a strong smart grid.Users can adjust their electricity consumption behavior according to their own load mode.This dissertation is mainly based on the original data provided by uk-dale,an open source data set in the UK.Firstly,the target data is extracted from the power load data provided by the data set.Then,the target data to be studied is preprocessed,and the characteristic vectors which can represent the daily load curve are obtained,Gauss mixture clustering and K-means clustering are used for clustering division.Finally,a load forecasting model is built for neural network forecasting.The main research results are as follows:(1)For the original power load data,in order to get high quality and reliable data information,data preprocessing is needed.In this paper,periodic analysis is used to fill the missing values,3 σ and quartile outliers are removed,and normalization based on min max standardization is used to process the original data.(2)On the basis of considering the characteristics of power load curve and the law of power signal,the feature decomposition vector based on VMD sample entropy and the feature vector of multi time domain statistics are further obtained to complete the two dimension feature vector extraction methods,which can be used for load pattern clustering and load forecasting on the basis of typical load patterns.(3)Aiming at the existing power load data,this paper divides the load mode,studies the analysis method of using Gaussian mixture clustering model and K-means clustering model,and combines the two eigenvectors respectively to obtain four mode division models.According to the analysis of evaluation index,the feature vectors based on multi time domain statistics describe the information of power load samples more comprehensively,and the sample pattern division based on time domain statistics Gaussian mixture clustering is more accurate and effective.(4)In order to monitor the applicability of power load model in real time,based on the division of typical load mode,the load mode with the largest number of samples is selected to forecast the load at a certain time point.Wavelet neural network and Elman neural network are used to compare and extract the power load samples under each load mode,which are divided into training samples and evaluation samples.Combined with the multi index evaluation,it shows that the wavelet neural network forecasting model is more suitable for the corresponding power load forecasting model.