Applications of neural network to short-term electric load forecasting
Xin Lin · Shenyang Gongye Daxue xuebao · 2006
The constituting process and training method of the improved BP and RBF neural networks are put forward.In the improved BP network,the momentum item and the algorithm using variable step length are employed.Furthermore,main meteorological factors influencing load changes are included in proposed mathematical model to meet weather variations.In the RBF network,to overcome the defects of traditional K-means scheme with local search,an orthogonal least square algorithm is used to select RBF center.By the improved BP and RBF neural networks,short-term electric load is forecast and training(convergence) rate and forecasting precision are analyzed.Comparing with BP network,the RBF has more advantages in practical applications.