Quality Prediction Based on Convolutional Neural Network

In modern chemical processes, quality prediction is essential for continuous characterization of dynamic behavior of chemical process. Traditional measurements for production quality have hysteresis, and have disadvantages of long processing time and difficulties in processing large amounts of data. Over the past few years, convolutional neural network (CNN) has demonstrated excellent performance in big data processing tasks, able to learn the characteristics of a large amount of input data more quickly and accurately, and to overcome the shortcomings of traditional prediction methods. But it was originated and still has been mainly used for classification issues. In order to establish a quality prediction model for non-linear process with higher accuracy, based on a study of the non-linear feature extraction ability of CNN, we tried to use it for regression issue for the first time and applied it on the quality prediction of Tennessee Eastman (TE) process. At the same time, compared with other prediction methods, the prediction effect of CNN is proved. It can be seen from results: (1) CNN can output predicted value in a short time, and according to root mean square error (RMSE) of network output, prediction of CNN is more accurate. (2) CNN uses fewer parameters and can help people save computer operating costs; (3) CNN can capture internal characteristics of data, saving people the time to extract the correlation of variables.