Application of a Combined Prediction Method Based on Temporal Decomposition and Convolutional Neural Networks for the Prediction of Consumption in Polysilicon Reduction Furnaces

Countries all over the world are making their contribution to common goal of energy saving and emission reduction. Solar is gaining more attention as a renewable source. Polysilicon an important raw material for solar panels production polysilicon vital part photovoltaic industry. typical process industry enterprise, its continuous highly intensive. Therefore, it necessary forecast analyze consumption plants. To address problem that difficult predict future based on historical data alone due time-series, massive, nonlinear, complex nature in workshops. This study proposes combined workshop prediction model Bayesian estimation time-series decomposition convolutional neural network (TSD-CNN). The method uses periodic long-term trend components algorithm optimally estimate parameters each component. With real-time collection equipment, application described above has successfully improved accuracy prediction, management efficiency safety warning capabilities enterprises. Furthermore, helps provide decision support conservation reduction studies. verify practicality reliability practical applications, experiments were conducted with furnace plant, by comparing commonly used regression methods SVM, LSTM TSD, results show greater improvement prediction.