Research and Application of Process Object Modeling Method Based on Deep Learning

Abstract

Predictive control is often used to improve control performance because most industrial sites are inertial lag objects, but it is difficult to establish predictive model. In this paper, artificial intelligence method is applied to control object modeling. A deep learning based modeling method is proposed, which includes open-loop modeling and closed-loop modeling. Closed-loop modeling is based on open-loop modeling. Firstly, the DNN deep learning algorithm used in this paper is introduced. Aiming at open-loop data of multi-order inertial system, a deep learning DNN network system based on Multiple Inertial filters is designed. After training step disturbance data, a deep learning neural network model with object characteristics is obtained. Secondly, a closed-loop modeling method based on two DNN models is proposed for multi-order inertial closed-loop systems. Firstly, forward and backward step disturbances are added to the control variables, then model 1 is obtained after training the DNN network system with deep learning based on Multiple Inertial filters, and then forward and backward step disturbances are added to the set values. The output data of model 1 and the output data of controlled objects are trained as inputs of deep learning DNN random inactivation network, and the closed-loop object model is finally obtained.