A temperature field prediction method for aluminum electrolysis cell based on improved CNN-LSTM

In traditional aluminum electrolysis processes, temperature measurement is often both time-consuming and labor-intensive, and can be significantly affected by environmental interference. To address these challenges, this study proposes a temperature field prediction method for aluminum electrolysis cells that integrates Proper Orthogonal Decomposition (POD) with Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM). Firstly, the POD method is used to reduce the dimension of complex temperature field data and extract key modal information, so as to effectively simplify the data dimension while retaining the core features. Second, the local feature extraction ability of CNN and the advantage of LSTM network in time series prediction are utilized to construct a CNN-LSTM hybrid prediction model. Finally, to verify the performance of the model, it is applied to the rapid prediction of aluminum electrolytic cell temperature field and compared with traditional measurement methods. The results show that the proposed model can accurately capture the dynamic trend of temperature field data. Compared with finite element analysis, the model has faster prediction speed and higher efficiency, with a maximum relative error of only 0.84% and an average relative error of only 0.06%. These findings confirm the stability and reliability of the model in practical applications and provide a novel and efficient technical approach for real-time monitoring of aluminum electrolytic cell temperature.