A CNN Model Predictive Control Approach Based On Time Delay Analysis Of Process Correlated Variables

In this paper,a prediction model based on convolutional neural network(CNN)combining the time delay analysis of correlated variables for production processes data-driven control is proposed,which is applied to predictive control.The main research work is presented as follows.Firstly,a prediction model based on CNN combining the time delay analysis of correlated variables for production processes is proposed.Time delays between the variables and the time series output are estimated,which are employed to result in the size of temporal windows of the associated variables.Then appropriate CNN models are established to demonstrate the effectiveness of the proposed method.Secondly,an improved CNN structure is proposed to adaptive model prediction control.The input and output of the network are re-select,and the control variables are added to the input.A CNN structure which can predict multi steps is proposed to improve the real-time of the algorithm.Then PSO is used for rolling optimization,and the prediction model is adjusted real-time by feedback correction to achieve prediction control.Finally,a reactor and distillation process experiment is carried out,which shows that the proposed method can enjoy good accuracy and robustness processes control.We use Caffe deep learning framework to train and test the CNN model,demonstrating the effectiveness of the proposed method.