Research And Application Of Power System Load Forecasting Based On Ridge Regression And LMS

The plan of power grid is the foundation for construction of power grid,and power system load forecasting is the basis of power grid planning.Thepower system load forecasting of power system is also an important part of the distribution system.It is regarded as the prerequisite of realizing the automation of power generation and economic dispatch.It is also an important basis for ensuring the security,reliability,stability and economy of the power grid management system.Meanwhile,with the urgent requirement of speeding up construction for “Resource Saving and Environment-friendly society,thepower system load forecasting of power system is playing an increasingly important role.Based on an analysis about background and significance of thepower system load forecasting,this paper elaborates the existing problems in the medium and long-term loading forecast of power system based on the current research status.The problems of missing data filling,data dimension reduction,multiple common linear identification and processing,discrete data identification and processing are analyzed and studied along the medium and long term loading forecast of the power system.The specific work is as follows:Based on thepower system load forecasting in power system,uncertain system of small data is often used as the research object.The model OGM(1,N)is improved based on GM(1,N).By discovering partial known information and extracting valuable information,operational behavior of system can be forecast.Because of the disadvantage of the value of the background value coefficient,a hybrid algorithm CPG based on chaotic mapping is proposed.the convergence of CPG is studied,the sufficient condition for the convergence of the algorithm is given and proved.In the end,an improved OGM(1,N)model for optimizing the background value coefficient is obtained,which is COGM(1,N)model.Due to the changes of recording personnel and mistake in data storage department,the sample data is often partially missing.Due to thepower system load forecasting in power system based on small sample model,the loss of modeling data,especially,the missing data at the back of sample sets can result in bad forecast for development trend of power load by forecasting model.This paper use three spline interpolation method to fill missing data based on the analysis of traditional filling methods for missing data.The existence of multiple collinearity in the sample data often has a great influence on the accuracy of the forecasting model.This paper tests the multiple collinearity in the statistical data of the production and operation of Jiangsu power grid.It is proved that there is a real multiple collinearity problem in the sample data.To solve this problem,a multiple collinearity solution based on stepwise regression and ridge regression method is proposed and applied to the OGM(1,N)model.The simulation results show that the OGM(1,N)model using the stepwise ridge regression method not only plays the role of signal dimension reduction,but also the stability is the highest in the proposed multiple collinear processing model.Sample data is the basis of load forecasting,and whether the accuracy of the sample data will directly affect the accuracy of the forecasting results.Normally the value of load is obviously higher than its value of historical trend called discrete data.Based on the analysis of the reasons resulted by discrete data,the identification of discrete data and the disadvantages of the traditional processing method for discrete data,this paper proposes an adaptive filter using the minimum mean square estimation(LMS)as the core,and applies it to the OGM(1,N)model.The simulation results show that the OGM(1,N)model with adaptive filter has strong robustness on the interference of discrete data.The proposed model is applied to the future planning of Fuxin power network as an application example of the proposed model.The rationality of power network planning is verified by power flow calculation,and some suggestions are put forward by short-circuit current calculation.