Research Of Combination Forecasting Model In Load Forecasting Of Power Grid

Power system short-term load forecasting(STLF) is an important task of power utilities, so great attention has always been paid on methods of STLF. Load forecasting is related to operation security and economical dispatching of power system,which is used to arrange the equipments dispatching and repairing .Also it can advance the stability of power system and save generation costs. With the development of area power market in China, STLF will play an important role in the operation of power market.There are many methods for STLF nowadays, but there hasn't a method which can obtain the satisfied forecasting results on any occasion for the reason of lack of accuracy. Linear combination and variable weighting combination forecasting model has been made,used single exponent smoothness model improve embodying continuous variance based on the model of artificial neural network, according to the law of power load variance in this paper. Through research of the two models and compared to simulation among traditional single forecasting model, it comes to the conclusion that the forecasting model and strategy in this paper can receive more precise result.The paper through the research of l load forecasting mode, combined with situation and characteristics of load features in GuangZhou area, mainly to do some work:1,Introduce the concept and rationale of power system load forecasting, meaning and purpose of STLF, some of the main method of study and application about the STLF at home and abroad, Listing several methods to normalizing historical data .2,Introduce the principles, structures and basic characteristicsthe of model of artificial neural network, focuses on Structures, algorithms and application in STLF of Forward neural network.3,Introduce linear combination and variable weighting combination forecasting model. For example, compared to simulation for the actual load of GuangZhou power grid, it proved that the superiority of combination forecasting model compared with single forecasting model.