ADAPTIVE PREDICTIVE PI CONTROL BASED ON NEURAL NETWORKS FOR ELECTROHYDRAULIC FORCE SERVO SYSTEM

An adaptive predictive PI controllor based on neural networks for electrohydraulic servo system is presented with respect to its complex nonlinearities and uncertainties.The method can be realized by a parallel structure which is composed of a real-time controller for real-time control operation,a predictive controller for regulation of control parameters and a supervising mechanism to determine whether or not the realtime controller is replaced by the predictive controller according to the variations of control performance. Thus, the real-time controller can be adjusted independently of the real operation of the controlled system by introducing neural networks as predictive model and predictive controller.It avoids the difficulty of the real-time control in control methods based on neural networks,thus makes the on-line learning control of complex systems possible.The learning ability of neural networks enables the predictive model to be modified on-line to simulate the behavior of the controlled system, and makes the predictive controller adjustable adaptively to variation of the system dynamical properties.Simulation results demonstrates its satisfactory control performance and it has been applied to a electrohydraulic servo structural testing system with success.