Time series prediction by growing lateral delay neural networks

Time-series prediction and forecasting is much used in engineering, science and economics. Neural networks are often used for this type of problems. However, the design of these networks requires much experience and understanding to obtain useful results. In this paper, an evolutionary computing based innovative technique to grow network architecture is developed to simplify the task of time-series prediction. An efficient training algorithm for this network is also given to take advantage of the network design. This network is not restricted to time-series prediction and can also be used for modelling dynamic systems