Short-term load forecasting using bayesian neural networks learned by Hybrid Monte Carlo algorithm

This paper presents a short term load forecasting model based on Bayesian neural network (shorted as BNN) learned by the Hybrid Monte Carlo (shorted as HMC) algorithm. The weight vector parameter of the Bayesian neural network is a multi-dimensional random variable. In learning process, the Bayesian neural network is considered as a special Hamiltonian dynamical system, and the weights vector as the system position variable. The HMC algorithm is used to learn the weight vector parameter with respect to Normal prior distribution and Cauchy prior distribution, respectively. The Bayesian neural networks learned by Laplace algorithm and HMC algorithm and the artificial neural network (ANN) learned by the BP algorithm were used to forecast the hourly load of 25 days of April (Spring), August (Summer), October (Autumn) and January (Winter), respectively. The roots mean squared error (RMSE) and the mean absolute percent errors (MAPE) were used to measured the forecasting performance. The experimental result shows that the BNNs learned by HMC algorithm have far better performance than the BNN learned by Laplace algorithm and the neural network learned BP algorithm and the BNN learned by HMC has powerful generalizing capability, it can welly solve the overfitting problem.

[1]  Carlos E. Pedreira,et al.  Neural networks for short-term load forecasting: a review and evaluation , 2001 .

[2]  Radford M. Neal.  Bayesian training of backpropagation networks by the hybrid Monte-Carlo method , 1992 .

[3]  G. Harrison,et al.  Analyzing the Impact of Climate Change on Future Electricity Demand in Thailand , 2008, IEEE Transactions on Power Systems.

[4]  Kotaro Hirasawa,et al.  A new learning method using prior information of neural networks , 2007, Science in China Series F: Information Sciences.

[5]  S. Duane,et al.  Hybrid Monte Carlo , 1987 .

[6]  Christopher M. Bishop,et al.  Pattern Recognition and Machine Learning (Information Science and Statistics) , 2006 .

[7]  David Mackay,et al.  Probable networks and plausible predictions - a review of practical Bayesian methods for supervised neural networks , 1995 .

[8]  D.W. Bunn,et al.  Forecasting loads and prices in competitive power markets , 2000, Proceedings of the IEEE.

[9]  Kotaro Hirasawa,et al.  A New Learning Method Using Prior Information of Neural Networks , 2000 .

[10]  R. Buizza,et al.  Using weather ensemble predictions in electricity demand forecasting , 2003 .

[11]  Nasser M. Nasrabadi,et al.  Pattern Recognition and Machine Learning , 2006, Technometrics.

[12]  LUeBaiquan,et al.  A new learning method using prior information of neural networks , 2004 .

[13]  R. Adapa,et al.  The impacts of temperature forecast uncertainty on Bayesian load forecasting , 1998 .

[14]  Kotaro Hirasawa,et al.  A new learning method using prior information of neural networks , 2006, Artificial Life and Robotics.

[15]  Geoffrey E. Hinton,et al.  Bayesian Learning for Neural Networks , 1995 .

[16]  George G. Karady,et al.  Advancement in the application of neural networks for short-term load forecasting , 1992 .

[17]  David J. C. MacKay,et al.  A Practical Bayesian Framework for Backpropagation Networks , 1992, Neural Computation.

[18]  M. Stone.  Cross‐Validatory Choice and Assessment of Statistical Predictions , 1976 .

[19]  Wray L. Buntine,et al.  Bayesian Back-Propagation , 1991, Complex Syst..