Direct Inverse Neural Network Control of A Continuous Stirred Tank Reactor (CSTR)

In recent years, there has been a significant increase in the number of control system techniques that are based on nonlinear concepts. One such method is the nonlinear inverse model based control strategy. This method is dependent on the availability of the inverse of the system model. Since neural networks have the ability to model any nonlinear system including their inverses and their use in this control scheme is promising. In the present paper, direct inverse neural network control strategy for controlling the CSTR with van de vusse reaction is studied. The direct inverse NN control strategy utilizes the process inverse model as controller. For training the neural network, the process input-output data is generated by applying a pseudo random signal on a simulink model of the CSTR process. Then, the input-output data is divided into two parts for training & validation. Training is performed using the Levenberg-Marquardt method. Based on the SSE, the optimum number of hidden nodes is taken as ten. The model obtained after the training is inverse NN model, which is taken as NN based controller. The performance of proposed NN based controller is evaluated for servo and regulatory control problems through simulation studies. Through the closed loop simulation studies, it is found that neural network based direct inverse control strategy gives superior performance to PID controller for setpoint changes. To improve the performance of direct inverse NN controller for regulatory problem, the IMC structured with forward and inverse NN models are included in the closed loop system.