A compound Bayesian networks gas prediction and scheduling method for blast furnace systems under various scenarios

Changes in the production scenarios of the blast furnace system disrupt the established production rhythm and gas balance. The complex characteristics of equipment in different scenarios and the temporal coupling resulting from the interactions between scenarios present challenges in prediction and scheduling. To address this issue, a compound Bayesian networks (CBN) gas prediction and scheduling method is proposed. In light of the characteristics of gas production and consumption data from different equipment, a monotonic variable-step size sliding window event extraction algorithm is designed to extract event information and construct state and delay event sets. By integrating the physical production structure with these event sets, state Bayesian networks, delay Bayesian networks, and an inter-network state representation layer are constructed, forming a complete CBN. Meanwhile, a prediction and scheduling inference procedure with temporal decoupling is proposed. It decouples the energy states of production equipment across successive scenarios and incorporates the objective function to enable scheduling inference and decision-making. Actual data from the steel production process are used for experimental analysis. The results demonstrate that the proposed method enhances prediction accuracy by 4.93% and reduces delay by 5.97%. In scheduling, it decreases the proportion of instances exceeding the safety tank level by 16.67% and lowers the comprehensive energy consumption per ton of steel by 0.0054%.