Research Of The New Methods Of Very Short-term Load Forecasting

Along with our country's electric power industry developing and the electrical network's expansive, the structure and the movement way of the electrical network are becoming more and more complex. In the course of electric power marketing, the accuracy of power load forecasting is directly related to the interests of all parties. The sudden increasement or the reduction both can bring bad influence to the electrical power system safe operation. It is the necessory to forecast the load effectively on line.Very short-term load forecasting means forecasting the load in an hour. It is very important to the control, the operation of the electrical power system. Improving its accuracy can strengthen the security and the efficiency of the electrical power system.This paper mainly analysises the Algorithmic structure, arithmetic speed, accuracy and error of tranditional methods, the three new very short-term load forecasting methods are studied.The first one is very short-term load forecasting method based on the local shape similarity. This mothod defines load curve shape coefficient based on analysis of power load local characteristic. It emphasizes on that the value forecasting should base on shape similarity. The influence of smoothing effect on inflexion point load forecasting caused by using homology weights commonly used in the existing forecasting methods is overcome.The second one is shape similarity criteyion based curve fitting algorithm. The criterion ensures the shape of the fitted curve should be most similar to that of actual curve is abided and at the same time the time-concerning impact factor is inducted, so the solution of parameters of fitting curve equation is transformed into constrained minimization problem. The proposed algorithm improves the curve fitting forecasting method in ultra-short-term load forecasting.The last one is time-varying nonlinear power load combined forecasting algorithm based on meta-learning. Meta knowledge formed by the results of base predictors and feature attributes of series is used as inputs of meta predictor when combined forecasting is applied. System bias can be founded and rectified. at the same time, the weights of base predictors are calculated using gating network. Results show that the first two method can improve forecasting precision including inflexion point, the third one can rectify system bias and improve forecasting accuracy in very short-term load forecasting.