A dynamic artificial neural network model for forecasting nonlinear processes

This paper presents the development of a dynamic architecture for artificial neural network (DAN2) model for solving nonlinear forecasting and pattern recognition problems. DAN2 is a data driven, feed forward, multilayer, dynamic architecture that is based on the principle of learning and accumulating knowledge at each layer and propagating and adjusting this knowledge forward to the next layer. Model building is automatically and dynamically repeated until a model that accurately captures the behavior of the process is determined. The resulting model is then used to forecast future values. To assess DAN2’s effectiveness, we present forecasting results for a variety of nonlinear processes that have been extensively studied in the literature and report comparative results. The set of nonlinear processes considered covers most nonlinear formulations facing researchers. We show DAN2 to be more accurate and to perform consistently better than alternative approaches employed in forecasting nonlinear processes.