Bayesian - BP Neural Network based Short-term Load Forecasting for power system

Short-Term Load Forecasting (STLF) is a very important aspect of power system to ensure operating safely economically and achieve scientific management in the power system. In this paper, Bayesian - BP Neural Network model has been designed for STLF. We used Bayesian - BP Neural Network to forecast the hour power load of weekdays and weekends. For doing this, Bayesian learning method has been used. This type of learning enables us to obtain the most probablic values of hyper-parameters so as to get a optimal BP Neural Network architecture. The training and testing data were collected from the historical load data of metritorious power of some area power system in Guizhou province. The test results showed that the forecasting results of the Bayesian - BP Neural Network model are more close to their real values than that of other classic BP Neural Network model, and Bayesian-BP Neural Network can be very effective in overcoming the limitation of poor generalization in conventional back-propogation(BP) algorithms compared with others.