Abstract: A short term load forecasting model based on Bayesian neural network learned by the Hybrid Monte Carlo (HMC) algorithm is presented in this paper. The weight vector parameter of the Bayesian neural network is considered as multi-dimensional random variables. Using the weather factors and load recorders in training set, HMC algorithm is used to learn the weight vector parameter with respect to normal prior distribution and Cauchy prior distribution respectively. Two Bayesian neural networks learned by Laplace algorithm and HMC algorithm and the artificial neural network learned by the BP algorithm are used to forecast the hourly load of 25 days of April(spring), August(summer), October(autumn) and January(winter) respectively. There are eleven nodes in input layer, ten nodes representing the ten weather factor variables of current hour and the previous hour and one hour variable. There is one node in output layer, corresponding to the load on each hour. The experimental result shows that the roots mean squared error (RMSE) and the mean absolute percent errors (MAPE) of the Bayesian neural network learned by hybrid Monte Carlo algorithm both are much smaller than those of the neural networks learned by Laplace algorithm and BP algorithm. Hence, the forecasting model based on BNN learned by the HMC algorithm has higher forecasting precision, and can be used to short-term load forecasting.

Key words: Bayesian neural network, short term load forecasting, Monte Carlo algorithm, prior probability distribution, Hamiltonian dynamical system