Short-Term Power Load Forecasting Based On Probability Analysis And Improved Extreme Learning Machine

With the development of the national economy,the power industry is facing increasing challenges.Load forecasting is of great significance to the entire power industry.In order to achieve accurate load forecasting,especially in the short-term load forecasting,it is of great significance to improve the operational efficiency of the power operation entities and to enhance the reliability and safety of the grid operation.This article studies the short-term power load forecasting based on probability analysis and improved extreme learning machine.Firstly,the paper introduces the classification of power load,expounds the characteristics of power load change,analyzes the many external factors that affect the load change.By analyzing the load and meteorological data of a city in Zhe Jiang Province,the characteristics of the load were analyzed.Secondly,for single-variable load forecasting,a similar-day method is introduced,and k-Nearest Neighbor(k-NN)learning theory is introduced into the selection of similarity days,and combines the mean influence value algorithm to select the dominant factors that affect the load change.The selected weather factors,the type of the week and other variables are combined to calculate the Euclidean distance between each historical day and the similar day.The historical day with the Euclidean distance less than the threshold is selected as the similar day.The load data of a city in Zhe Jiang Province were simulated and analyzed to verify the rationality of the method in selecting similar days.The paper proposed a forecasting method of load probabilistic analysis based on similarity day combined with quantile regression of neural network.The nonparametric kernel density estimation theory and the neural network quantile regression model are introduced.By using the proposed method and the normal distribution and kernel density estimation methods,the load data measured in a certain city in Zhe Jiang Province are predicted respectively.And the results of point prediction and interval estimation are used to verify the feasibility and effectiveness of the proposed method.Then,according to the characteristics of multivariate weather factor data andinformation mutual embedding,the kernel principal component analysis is used to reduce the dimension of multivariate weather factor data to simplify the data structure.By discussing and analyzing parameters of different kernel functions,and comparing the obtained results with the results of principal component analysis,the mixed kernel KPCA is determined as a suitable method for processing multivariate weather factor data.This paper introduces an improved extreme learning machine with self-adaptive differential evolution algorithm to establish a predictive model for multivariate data,and the performance of E-ELM,DE-LM and Sa DE-ELM are compared,in the end,simulation analysis validates the effectiveness of the KPCA-Sa DE-ELM model.