Hybrid adaptive learning control of nonlinear system

A hybrid adaptive learning control for nonlinear dynamical systems is proposed. Feedforward multilayer neural networks are used to construct a controller. Parameters of the neural networks are adjusted by a dynamic backpropagation algorithm and a genetic algorithm. The genetic algorithm manages to escape local minima and reach the neighborhood of the global minimum on the squared error surface. The dynamic backpropagation algorithm is used to search the global minimum from its neighborhood. Computer simulations show that the tracking control performance of nonlinear dynamical systems can be enhanced by the proposed method.