Study On Electricity Forecasting Of Extreme Learning Machine With Generalized Correneropy Criterion Limimt Considering User Characteristics

blectricity forecasting technology is crucial in both the State Grid Corporation's bid valuation and electricity market nowadays.However,in the current electricity forecasting problems faced by the electricity selling companies the error distribution of historical electricity data and electricity forecasting is usually non-linear,non Gaussian distribution,and due to the small scale of users,some random events and data collection errors will lead to outliers.which has a great impact on the forecasting results which makes the traditional forecasting method no longer applicable.As a nonlinear forecasting model,the Extrene Learning Machine(ELM)has been proved to have good performance in the field of regression forecasting.However,the traditional ELM with Mean Squared Error(MSE)is more sensitive to outliers thus its output results are greatly affectcd by outliers,which makes it unable to correctly represent the error statistics information in the non Gaussian case in the electricity forecasting problems.In order to solve this problem,a new robust forecasting model is constructed by introducing Generalized Maximum Correntropy Criterion(GMCC).From the perspcetive of information theory GMCC as a new cost function can be used to solve the problem of non Gaussian signal procesing.Therefore,this paper use GMCC as new cost function to replace the MSE cost function in the original ELM,and contructs the GMCCELM electricity forecasting model.In addition.in order to solve the poblem that the left weight of GMCCELM network structure is given randomly and the output result is unstable,the Kernel Extreme Learning Machine(KELM)is introduced and the electric quantity forecasting model of GMCCKELM is obtained through improvement.In addition,through the method of data statistics and analysis,this paper studies the influencing factors of the user's electricity consumption in this case,and finds that the historical temperature and the historical electricity are the main factors affecting the future electricity change,so these two factors are taken as the input of the forecasting model.Finally,the historical electricity quantity of a commercial user in Guangzhou province and the corresponding temperature data are used to verify the practicability of the forecasting model established in this paper,and compared with other algorithms.The forecasting results show that,compared with the traditional ELM and other algorithms,the forecasting model established in this paper can effectively improve the accuracy of electricity forecasting and improve the robustness.