Analysis Of Electricity Consumption Behavior Of Consumers In Smart Grids And Prediction Of Power Load Probability Density

Electricity consumption forecasting is an important part of industry and energy sector macro planning.Accurate power load forecasting is critical to grid management and power dispatching.With the development of the power industry,scholars have conducted a lot of research and put forward various algorithms to improve the accuracy of power load forecasting.At present,the majority of the power load forecasting is based on regions.However,as the basic unit of electricity consumption,residents and small and medium-sized enterprises have the same importance in power load forecasting as regional.In addition,compared with the regional electricity consumption,the power consumption of residents and small and medium-sized enterprises is more uncertain and difficult to predict.Therefore,this study aims to combine the adaptive spectrum clustering(ASC)method with the support vector quantile regression model(SVQR)to analyze the power consumption behavior of smart grid users and forecast the power consumption of residents.Very-short-term power load forecasting is an online real-time load forecasting activity.The time interval for forecasting is generally within one hour.The time interval for load forecasting in this paper is half an hour.Very-short-term load forecasting can not only balance the power supply and demand,but also perform online control of the power grid andreasonably dispatch power generation capacity.In order to effectively quantify the uncertainty of the electricity load of smart grid users,this paper adopts electricity load data released by the Irish energy authority,and carries out point prediction,interval prediction and probability density prediction of residential electricity load by taking into account factors such as time,holiday and historical load.The evaluation indexes MAPE,PICP,PINAW,pinball loss and newly proposed MPPL in this paper were calculated to compare the prediction effect of the model.Due to the great difference in power consumption between winter and summer,this thesis considers load forecasting on winter and summer data sets respectively.By using the spectral clustering method,users with the same power consumption patterns are clustered into one cluster among the 5087 users.All users are thus grouped into six clusters.According to different user categories,SVQR model is used to make point prediction of power load,and compared with other algorithms such as single support vector regression.Here grid search is used to optimize the parameters of the model.By comparing the two error evaluation index values of MAPE and pinball loss for different models,it is shown that the SVQR model has the best prediction performance.In order to effectively provide the uncertainty information of power load,SVQR algorithm is adopted in this study to carry out the interval and probability density of the load of userswith ultra-high energy consumption and users with medium energy consumption at any time in the future day.In this paper,it is concluded that compared with other single support vector regression models,the prediction accuracy of SVQR model is higher.Besides,the prediction results of SVQR model still have high reliability in the case of relatively narrow bandwidth of prediction interval.What's more,the probability density prediction results of the very short-term load obtained by using SVQR model can provide comprehensive information about the uncertainty of future load.Since the selection of kernel function becomes very important when the support vector quantile regression model is used for prediction,this paper considers to compare two different kernel functions: Gaussian kernel function and Laplace kernel function.At the same time,in order to test the generalization ability of the model,a longer data set is selected for simulation.The results show that the quantile regression prediction effect of Laplace kernel support vector is better for ultra high energy consumption and high energy consumption users,while the quantile regression prediction effect of Gauss kernel support vector is better for other types of users.In general,the Gaussian kernel function has better prediction effect.To sum up,the method used in this paper can not only solve the nonlinear problem of the relationship between the influencing factors ofload forecasting,but also solve the uncertainty problem of load forecasting.