Hybrid self-learning model for the prediction and control of sintering furnace temperature

Ternary cathode materials are important components of lithium-ion batteries. However, the sintering process during manufacturing is challenging to control due to the inaccessibility of key dynamic variables and the frequent fluctuations in operating conditions. These lead to high energy consumption and inconsistent product quality. In this paper, we propose a hybrid self-learning prediction model and control method for sintering furnace temperature based on both first-principle and process data. Firstly, a mechanism model with temperature time delay is established based on energy flow analysis in the furnace. To capture the tail gas temperature dynamic in the mechanism model, a Ventingformer-based prediction data-driven model is proposed. In this model, a memory updating technique and an autoregressive module based on the Transformer framework are developed to identify long-time dependencies and respond to variations in input sequences. Then, a hybrid self-learning modeling framework is designed. Based on the established hybrid model, a multiscale objective function-based nonlinear model predictive control (MSCF-NMPC) method is proposed to achieve precise tracking control of the internal temperature in the furnace. A multiscale objective function with short-term cost in terms of energy consumption and tracking accuracy as well as long-term cost in terms of energy loss is constructing in the control optimization problem. Finally, the proposed hybrid self-learning model and MSCF-NMPC method are verified using the actual process data from a sintering furnace, demonstrating the effectiveness of the proposed method. The results offer practical guidance for industrial applications.