Deep Learning Strategy and Experimental Research of Kiln Temperature Based on Predictive Control

Abstract The roller hearth kiln (RHK) is a high-precision sintering equipment for semiconductor passive ceramic components. Its internal temperature control is a complex process of multi-input and multi-output, involving significant time delay, nonlinear, and multifactor spatial distribution characteristics. Most traditional control methods are based on feedback adjustment and lack of coupling information between variables, so it is difficult to meet the requirements of high-precision industrial control. Therefore, this paper proposes a deep learning strategy of kiln temperature based on predictive control for real-time temperature control of a roller kiln. To ensure the reliability of the control strategy, this study designed and built a kiln test platform according to the working principle, structure, and control characteristics of the roller kiln, and realized the functions of disturbance generation, communication, and distributed temperature control that mimic the actual scene. The results show that the model can accurately predict the dynamic change of temperature with time in the kiln temperature control system, and the prediction error of the model is controlled between 0.1 °C and 1 °C. By combining simulation and experiment, stable temperature control can be achieved under normal working conditions and disturbances. The final error range is only 0.1 °C∼0.2 °C, proving its effectiveness and reliability as a multi-input and multi-output temperature control strategy for the kiln.