Research Of Mid-long Term Power Load Forecasting Based On Grey Theory And Neural Network

Power load forecasting plays an important role in the stable and economical operation of power system. Mid-long term load forecasting provides data support for power system planning departments when they need to make decision. The accuracy of mid-long term load forecasting directly influences the rationality, economy of the power grid reformation and expansion, which has profound significance to the development of power industry.Because mid-long term load forecasting has long time span, and it is influenced by the economy, policy, population growth and other macroscopical factors, so its research work is very difficult.Viewing the existing mid-long term forecasting method, mid-long term load forecasting technology still has large research space and need to be further improved. The important research content of this subject is that analysising mid-long term load historical data rule deeply, considering the main factor which influences power load and determining the model which is suitable for mid-long term load forecasting.Firstly this paper elaborates the background and significance of mid-long term load forecasting, and analyzes the development of forecasting technology. Secondly this paper mainly narrates power load forecasting theory in detail, and discusses factors which influence power load forecasting. In view of mid-long term load forecasting whose feature is very consistent with the feature of grey forecasting, this paper uses grey model to forecast mid-long term load and studies the modeling mechanism of grey model deeply. In the aspect of selecting grey models, this paper uses the classical GM(1,1) model firstly. Because mid-long term load is influenced by many factors, this paper adopts GM(1,n) model and GM(0,n) model which consider related factors, and uses grey related analysis method to screen the main factor. Through an in-depth study of three grey models and analysis of forecasting results, this paper pointes out the limitations of three models. This paper deems that there is a function relation between the actual load value and forecasting results of three models. Because BP neural network has strong nonlinear mapping ability and good learning ability, in order to improve the forecasting results, the paper uses BP neural network to fit the function. By combining three grey models and BP neural network in a specific way, this paper designs GM-NNC model. Then this paper designs RGM-NNC model, which is on the basis of GM-NNC model and adds the idea of equal dimension and new information technique. RGM-NNC model improves unchanged historical data shortcoming of model, so the new data can be fully utilized. This paper uses Microsoft Visual C++ 6.0 to program two kinds of design models, and uses examples to test and verify the forecasting results of two models, and uses MATLAB2010 b to simulate the forecasting results. The results indicate that the forecasting accuracy of two new kinds of models has been significantly improved. The two new models are suitable for mid-long term load forecasting and have practical application value.