Application of Neural Networks and Kalman Algorithm in Oil Demand Prediction

This paper analyzes the eight indicators which affected oil demand from the year 1978 to 2009. Setting the indicators into three groups and applying different group data, China's oil demand is estimated in 2013 by the method of generalized regression neural network(GRNN) and back propagation neural network(BPNN). Five indicators which have great influence on oil demand are picked via Mean Impact Value method. Based on the five variables, the order of autoregressive(AR)model is established by employing Akaike information criterion. A posteriori estimation of the AR model is carried out by the Kalman algorithm and Rauch-Tung-Striebel(RTS) algorithm. The results indicate that Kalman filter algorithm could amend the AR model well by updating the parameters and improve the accuracy of the prediction.