Research On Analysis And Identification Of Power Consumption Behavior Patterns Of Industrial Users Based On Data Mining

In order to achieve the goal of "double carbon",China is accelerating the construction of new power system,resulting in the continuous increase of new energy power generation on the power generation side,the continuous change of power consumption mode on the power consumption side,and the deepening of information exchange between supply and demand sides.The analysis and identification of power user behavior patterns based on load big data is the basis of accurate control of real-time load,reasonable guidance of orderly power consumption and energy efficiency management,and is of great significance to the balance of power supply and demand.Therefore,how to effectively extract and accurately identify the behavior patterns of power users from the perspective of data is an urgent problem to be solved,this thesis will focus on industrial power users.In order to deeply analyze the power consumption behavior of industrial users,firstly,considering whether the user’s production mode is affected by the calendar effect,the dynamic time warping（DTW）similarity with high robustness to random fluctuation is introduced as the evaluation index to classify the load periodicity.Then,based on the generalized extreme value（GEV）distribution,the K-means algorithm is improved to cluster the periodic load.The clustering center is selected according to the global information of the load data sample to reduce the influence of boundary points and outliers,improve the clustering accuracy and verify its effectiveness.Finally,the load without obvious periodicity is fitted and modeled based on the long-term load time series.The variational mode decomposition（VMD）algorithm combined with the gated recurrent unit（GRU）neural network method is used to smooth the data and remove the redundant noise while retaining the complete load characteristic information and improve the accuracy of load modeling.For new enterprises without historical load data,it is impossible to analyze their power consumption behavior mode,so it is necessary to use the historical load information for reference.Taking the results of load clustering as the category label,the sparse representation classification（SRC）model based on the discriminant dictionary trains the dictionary and classifier model at the same time,and takes the sparse representation coefficient with the largest absolute value as the classification criterion to realize effective load type identification so as to take corresponding control measures.