Generalised Nonlinear Pid Controller Based on Neural Networks

In this paper, a nonlinear controller is proposed to handle some nonlinear control problems. In this scheme, the controller uses the system error, the integral of the system error, and the derivative of the system error as its inputs but the mapping from the inputs to the output is nonlinear. The corresponding nonlinear mapping may be specified based on the control requirement. Therefore, the proposed controller is defined as a generalized nonlinear PID controller (GNPIDC). The GNPIDC strategy is realized using neural networks. For on-line training of the neural network based GNPIDC, a PID gradient descent optimizing algorithm with momentum term (PIDGDM) is proposed. Then, the convergent characteristic of the algorithm is presented. Finally, simulation study of applying the neural GNPIDC strategy to a continuous-stirred-tankreactor (CSTR) and a van de Vusse reactor is illustrated.