Modeling and control of batch distillation process by neural network approach

Abstract

Batch distillation, one of separation processes used in many industries, especially food, pharmaceuticals, and fine chemicals, is inherently complex and nonlinear dynamic behavior and is therefore very attractive issues in determining reliable models and appropriate control systems. In this work, a multilayer feedforward neural network is applied for modeling and control of a batch distillation process. This work is divided into two sections: system modeling, and control system applications. In the first one, it can be seen that the network with two hidden layers is able to represent the process behavior better than that with only one layer. In the other one, a model predictive controller based on neural network models is formulated to control the cyclohexane composition at the top plate. It can be seen that the controller provides good control response for tracking the cyclohexane composition determined by an optimization approach of batch distillation process under an optimal reflux ratio condition. Moreover, the controller gives robust tracking capability under plant/model mismatch.