Digital Twin-Empowered Power Consumption Prediction for Energy-Intensive Aluminum Annealing Furnaces

The aluminum annealing furnace (AAF) is a large, energy-intensive industrial equipment widely used in manufacturing. Accurate prediction of its power consumption is crucial for optimizing energy management. However, conventional prediction methods often face challenges due to the furnace's multi-day production cycles, coupled operating conditions, and complex thermal interactions. To overcome these challenges, this paper proposes a digital twin (DT)-based approach for high-fidelity power consumption prediction of AAFs. First, a DT-empowered prediction framework is proposed, which establishes a closed-loop interaction between physical entities and virtual models through four synergistic layers: physical equipment, data integration, DT simulation, and application services. Within this framework, detailed power consumption profiles are generated via DT simulations that replicate the AAF's production process in advance, utilizing both the twin model and production data. The AAF DT model is developed using a CNN-BiLSTM-Attention network, effectively capturing the nonlinear dynamics of power consumption. Furthermore, an incremental learning strategy is implemented to continuously refine the model with increasing data, ensuring adaptability to varying production scenarios. Finally, a real-world case from an aluminum manufacturing facility is provided to validate the proposed approach. Experimental results demonstrate its superior performance, with all R2 values exceeding 0.96, consistently outperforming other baseline models.