Multi-Feature Fusion Prediction of Steelmaking by-Product Gas Based on Hierarchical Attention Actor-Critic Network

The effective prediction of energy from by-products in the steel industry is a prerequisite for optimizing industrial scheduling. Linz- Donawitz converter gas (LDG) is a significant secondary energy source. Due to the high production output, significant fluctuations, and diverse user groups of the LDG system, accurately predicting LDG generation flow presents considerable challenges. To address this issue, we propose a hierarchical attention multi feature fusion network to better integrate the features of multiple production variables. Methodologically, model matching method is used to extract time-domain and frequency-domain features that describe production rhythm. Subsequently, the proposed network will be established and formed into a dynamic prediction model using deep reinforcement learning. Finally, perform vector inverse reconstruction on the predicted rhythm features to obtain the time series prediction results. Simulation experiments were conducted using real data from a steel plant in China. The results demonstrate that the proposed method effectively integrates data from different sources and provides dynamically adjusted prediction results. Moreover, compared to other methods, it enhances prediction accuracy, thus offering scientifically effective guidance for subsequent energy scheduling.