Shortterm load interval forecasting based on gaussian fuzzy information granulation and improved wavelet neural network

YU Peng1,2, TANG Quan3, ZHANG Wentao3, HUANG Minxiang1

(1.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China;
2.State Grid Jiangsu Electric Power Corporation Economic and Technology Research Institute, Nanjing 210008, China;
3.State Grid Sichuan Electric Power Corporation Economic and Technology Research Institute, Chengdu 610041, China)

Abstract: Aiming at the problem of being lack of adaptability and forecasting accuracy of the current shortterm load forecasting methods, and the problem of slow evolution, easy to fall into a local minimum and no convergence because of the random assignment of the original connection weights and threshold and adopting gradient learning algorithm to improve the correction in original WNN method, a new method based on gaussian fuzzy information granulation and improved wavelet neural network was proposed. A faster convergence speed function was proposed to replace the commonly used output layer neuron function and particle swarm algorithm optimization was used to replace the random assignment of the original connection weights and threshold of WNN. The network connection weights and threshold was taken as a particle position vector of particle swarm optimization, the speed of the particles and the position vector was adjusted constantly to find the optimal value. Appropriate data span was selected as a graining window to dispose the original load data by using gaussian fuzzy information granulation method and the corresponding sequence values after gaussian FIG were gotten. Then the improved WNN was used to do interval forecasting of the fuzzy sequence values. The results through comparative study of WNN and SVM indicate that the method proposed can not only gain range information, which is more than single load value, but also achieves higher prediction accuracy, which can better guide the power system related decisionmaking.

Key words: gaussian fuzzy; information granulation; improved WNN; shortterm load; interval forecasting