Multi-step-ahead prediction using dynamic recurrent neural networks

A method for the development of empirical predictive models for complex pro cesses is presented. The models are capable of performing accurate multi-st ep-ahead (MS) predictions, while maintaining acceptable single-step-ahead ( SS) prediction accuracy. Such predictors find applications in model predict ive controllers and in fault diagnosis systems. The proposed method makes u se of dynamic recurrent neural networks in the form of a nonlinear infinite impulse response (IIR) filter. A learning algorithm is presented, which is based on a dynamic gradient descent approach. The effectiveness of the method for accurate MS prediction is tested on an artificial problem and on a complex, open-loop unstable process. Comparativ e results are presented with polynomial Nonlinear AutoRegressive with eXoge neous (NARX) predictors, and with recurrent networks trained using teacher forcing. Validation studies indicate that excellent generalization is obtai ned for the range of operational dynamics studied. The research demonstrate s that the proposed network architecture and the associated learning algori thm are quite effective in modeling the dynamics of complex processes and p erforming accurate MS predictions. (C) 2000 Elsevier Science Ltd. All right s reserved.