Adaptive prediction using neural networks

A novel adaptive multistep predictor based on backpropagation neural networks is developed for nonlinear dynamical systems, and the prediction mechanism is analyzed. Two isomorphic neural networks are used together to implement the proposed predictor. One is called the learning network (LN), and the other is called the prediction network (PN). The weights of the two networks are adaptively adjusted so that past predictions closely match the observed data. These weights are used to generate future predictions. Simulation results demonstrate the effectiveness of the predictor