The Combination Forecasting Of Bayesian Model Averaging Base On MCMC And Using It In Energy Consumption Forecasting

Energy is the important substantial foundation for social development. Energy consumption prediction plays an important role in making an energy plan, contributing to the rationality of energy production and consumption. With the development of the economy and the industrial process, the social demand for energy will increase greatly. The resources, however, are relatively in shortage, making China confronted with a long-term energy supply crisis. Thus, the energy supply problem, once more, is put on the agenda. Therefore, it is of vital importance to use the scientific and rational energy consumption model to predict the energy consumption in the coming years precisely. And it can make a significant contribution to making rational economic developmental strategies and energy security strategies.With the limits of single models, the combination prediction models are applied to more and more occasions in the reality, improving the precision greatly. The key to the combination forecasting is to determine the weight of each individual forecasting model There are three common shortcomings in the combination prediction models: first, the subjective information has not been considered ; second, the correct predicted information from various prediction methods has not been adopted sufficiently;third. It don't consider the problem of models'uncertainty. Bayesian Model Averaging Combination Prediction uses posterior probability. as the weight to calculate the weighted average for all the possible single prediction models. Having considered all the possible single prediction models and set posterior probability. as the standard to estimate the models, It has overcome the shortcomings of the existent methods of calculating combination weight and tackled the problem of models'uncertainty.The key of applying Bayesian Model Averaging is to estimate the weight of each Single model accurately. In this paper , the weight is estimated with the Markov Chain Monte Carlo Methods which provide an effective approach to calculate the marginal likelihood function.. Theempirical study of energy consumption in China suggests that combination Forecasting Value, generated by MCMC method with expectation maximization algorithm (EM), Akaike Information Criterion(AIC) and Bayesian Information Criterions(BIC), shows high forecasting precision; The combination forecasting method of Bayesian model averaging has very good application values in dealing with the energy consumption systems, which somehow contains some uncertain combination modeling and predictions.