Neural network-based control strategy for the two-phase reactor of the Tennessee Eastman process

Tennessee Eastman plant-wide control has been an area of intense research and investigation over the past several years. The interest in the Tennessee Eastman plant-wide control problem can be attributed to the highly non-linear behavior of the plant, which subsequently makes the plant difficult to control. This project attempted to control the reactor of the plant using a neural network. The neural network was trained using unsteady state data obtained from an open loop model. The neural network was used both as a stand-alone feedforward controller and in conjunction with a cascade controller. The performances of both these control systems were then compared with a cascade control system proposed by McAvoy. The control system using the neural network in conjunction with cascade control provided results similar to the cascade control scheme proposed by McAvoy. The neural network feedforward controller in conjunction with McAvoy's control scheme for the reactor provided slightly better control action initially. However, the neural network feedforward control did not improve overall plant performance.