Power Load Classification Method Based On User Behavior Characteristics

New energy consumption has been a huge challenge through the development of a high proportion of renewable energy.When there is a consumption gap in the balance of new energy power,this aggregator can organize the controllable load to increase the load in response to the clearing period and promote the consumption of electricity.New energy power controllable load transfer call to order take precedence over provincial thermal power load.Therefore,analyzing users’ electricity consumption behavior,studying load classification and finding out controllable loads are of great significance for load aggregators to formulate trading categories and trading rules for controllable load resources to participate in electricity market and fully mobilize controllable load resources to participate in market transactions.In this thesis,the non-resident commercial load and industrial production load are taken as the research objects,with the goal of high efficiency,stability,safety and environmental protection.Based on the analysis of users’ electricity consumption behavior,the user load data is preprocessed by similarity ranking based on load characteristic indicators,and then the load data is processed by color mapping and image processing.The convolution neural network with double-branch input is used to cluster the data,and the mutual information maximization objective function is used to classify the load data effectively.The main research contents are as follows:(1)Consider multi-index analysis of typical load curve characteristics to obtain the behavior characteristics of the electricity industry for different types of load curves.On this basis,this thesis put forward the subjective and objective evaluation methods of power consumption behavior of users in different industries.It is to determine the relative influence and importance of different load characteristic indexes by means of paired comparison and corresponding weight coefficient is obtained according to this result.(2)When traditional clustering algorithms in classification of load data easy to produce insufficient load randomness classifying load data according to Improved OTSU algorithm.In order to obtain data image processing for load classification,it is Perform threshold classification in the image histogram and use the weighted average method to perform weighted pixelation on the load curve image.(3)In order to realize the load classification method based on invariant information clustering,this proposed a clustering method based on two-branch input convolutional neural network.Therefore,in order to obtain the load curve types in the data set,it use the mutual information maximization objective function is used to calculate the load curve characteristics of the clustering results.In addition,this taken commercial load data and industrial load data are examples to obtain the method verity and effectiveness,and it use the load classification method based on invariant information to comparison and analysis of simulation experiments.