Abstract
In the last decade, neural networks have been widely used in performing time series prediction [11,13,15]. Long term prediction is generally far more difficult than short term prediction, because of the difficulty in modeling the system dynamics far ahead in the future[14,17]. In this paper, we present a novel algorithm for training neural networks for use in long-term prediction. Our algorithm relies on the utilization of traditional time series analysis based on Box- Jenkins methodology to : 1) determine the appropriate neural network architecture; 2) select the inputs to the neural network;3) determine the appropriate lead time for updating the connection weights of the neural network during training. We demonstrate the effectiveness of the algorithm on some real world time series data as well as simulated time series data.