Time Series Forecasting in a CVD Reactor for Polysilicon Production

Bangwen Xi, Gang Xiong, Jun Yan, Zhen Shen, Yonggang Song, Xiong Liu, Sheng Liu

Abstract: Usually high-grade polysilicon production is obtained by the Chemical Vapor Deposition (CVD) process in the Siemens reactor. Predicting key parameters in the reaction can help increase production, which is a time series regression problem. We propose to use neural networks to predict the key variables of the CVD reactor. The performance measure for the model is the mean standard error (MSE). The experimental results show that the MSE of the target parameter prediction is 0.00612. Compared with Artificial Neural Networks(ANN), Prophet, and other algorithms, this algorithm achieved high accuracy and has been used in actual production. We believe the proposed approach is of great interest for the operation of the polysilicon CVD process.