This paper describes an approach for auto-tuning PID controllers using neural networks. Compared with conventional PID controllers, the realization of the proposed controller requires only measured input and output signals of a controlled object. To improve the learning convergence speed, the multiple extended Kalman algorithm (MEKA) is employed for the training of the networks system. In addition, the application of a high gain observer makes this controller capable of being operated under the measured output of a system corrupted by a strong measurement noise. The same as other self-tuning controllers, the proposed controller can tune the parameters of controller automatically. Furthermore, since the networks can provide a nonlinear control signal, this method is suitable for controlling objects containing nonlinear uncertainties, such as pneumatic cylinders, electrical motors, etc. Simulation and experimental results demonstrate the effectiveness of the proposed method for control performance and disturbance resistance.
抄録全体を表示