Monthly Electric Power Forecasting Algorithm Based On Maximum Correntropy Criterion

With the gradual advancement of the second power reform,power sales companies are increasingly eager to find high-precision,universal power forecasting algorithms.The power forecasting algorithm with high accuracy and universal adaptability can provide the most basic strategic guarantee for the power selling company,which is of great significance for the enterprise to enhance the core competitiveness of the market.However,the market for electricity sales in China has only just begun.The electricity forecasting algorithm for small and medium-sized users has not received much attention.Traditional electricity forecasting algorithms can no longer meet the market demand of electricity sales companies.At present,the power consumption of many small and medium-sized users is easily affected by the economic benefits of the company and local climate factors,making the electricity data provided by the users have obvious characteristics such as non-linearity,randomness,etc.,resulting in the traditional prediction algorithm is difficult to accurately predict,resulting in The huge loss of the power sales company.The traditional neural network algorithm or extreme learning machine algorithm needs to consider the data amount of the model in the power prediction,that is,it needs to provide a large amount of historical data sets for training,which limits the application of these two types of algorithms in this field.Based on the above problems,this paper focuses on solving the small-scale,non-linear and strong power-data prediction problem,and proposes using a suppor;t vector regression algorithm(MCC-SVR)based on the maximum correlation entropy criterion and a least-squares support vector based on the maximum correlation entropy criterion.Machine Algorithm(MCC-LSSVM).This paper first defines the concept of power forecasting and briefly introduces the model that is commonly used for power forecasting.Secondly,the data's predictability is analyzed by using Deng's correlation degree and the improved correlation analysis model to find out the two factors that have the greatest impact on electricity consumption forecast:economic factors and climate factors.Then,based on the statistical characteristics of the small sample size and non-uniform distribution of existing data,MCC-SVR and MCC-LSSVM models are formed using the maximum correlation entropy criterion instead of the risk function in the original model on the basic support vector machine model.Finally,in order to make the improved model better able to adapt to electricity forecast under various conditions and maintain the best learning ability,this paper uses the grid optimization method and cross validation method to optimize the model parameters to ensure that the parameters are universal.Sex.Finally choose the appropriate evaluation indicators to analyze the forecast results.At the end of this paper,the data of the education industry,mechanical manufacturing industry and other industries are taken as samples,and the electricity consumption is predicted with the economic development data and climate data.The results show that the improved scheme based on MCC is effective and the prediction error is less than 3%.