An evolving dynamic graph based model for aluminum smelting furnace temperature prediction

Real-time temperature monitoring in high-temperature smelting furnaces is often costly and delayed due to the short lifespan and thermal inertia of contact sensors like thermocouples. To solve this problem, a furnace temperature prediction model is proposed in this paper, which simultaneously accounts for both temporal and spatial correlations. Firstly, the spatial feature extraction module employs a meta-adjacency matrix evolution mechanism to dynamically adjust the graph structure and utilizes Graph Convolutional Networks (GCN) to extract spatial features from the nodes. Then, the time series prediction module leverages Long Short-Term Memory networks (LSTM) to model the temporal sequences of the extracted spatial features, thereby forecasting future values. Experimental results obtained from real-world aluminum smelting furnace data demonstrate that the proposed model achieves superior prediction accuracy compared to LSTM and Spatio-Temporal Graph Convolutional Network (STGCN), with RMSE reduced by 60.2% and 55.0%, and MAE reduced by 52.3% and 43.5%, respectively establishing it as a robust solution for real time temperature prediction in aluminum smelting furnaces.