Temperature Prediction Study of Graphite Purification Zone in Arc Plasma Based on Intelligent Algorithm

Abstract

Predicting the temperature in the graphite purification zone of arc plasma is not only useful for guiding the smoothness of the production process, which affects the final quality of the graphite product, but also provides decision support for the implementation of fine temperature control. This study conducted an analysis of nine variables associated with the temperature of the graphite purification zone within arc plasma to develop temperature prediction models. Using the random forest algorithm for feature selection, four key variables were identified with importance exceeding 0.4. These variables include the main gas flow rate, powder flow rate, voltage, and current. Subsequently, three different algorithms, namely Error Back Propagation (BP), Extreme Learning Machine (ELM), and Long Short-Term Memory (LSTM), were employed to develop various models for predicting temperature. Ultimately, the predictive performance of the models was assessed by comparing the temperature prediction models for various operating scenarios using different evaluation criteria. The experimental results show that the three models have their own advantages and disadvantages in different working conditions. The LSTM model exhibited  superior predictive performance in C1~C2 and C6~C9 working conditions, the ELM model demonstrated superior predictive performance in C3~C4 working conditions, and the BP model demonstrated superior predictive performance in C5 working conditions. Hence, in the actual purification process, a variety of prediction models can be used in combination to facilitate the continuous monitoring of temperature variations in the graphite purification zone.