MA Li xin, ZHOU Shang jun xi

(School of Optical Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China)

Abstract: Aiming at the problem of the factors affecting maximum load in continuous multi days are complex, hard to forecast the multi days load, multi days load including holidays forecasting has low accuracy, characteristics of power load in recent years were analyzed, the Self Organizing Feature Map cluster method was used in data processing, loading data, characteristics of holidays were studied, a new method of multi days load forecasting base on Self Organizing Feature Map(SOM) neural network was proposed. The characteristics of the holiday and the non holiday daily peak load were investigated respectively. For the non holiday, the Elman neural network uses the training data which was selected by SOM clustering to forecast. For the holiday, the holiday factor is added. The experimental results show that the prediction accuracy of this method can meet the industry requirement, the average error is 3.21%.It provides a feasible way for multi day load forecasting.

Key words: multi days load forecasting; feature extraction; Self Organizing Feature Map clustering; maximum load