A robust optimization method for multi-cast batching plans and casting start time dynamic decision in continuous casting process

Abstract

The cast batching planning and the decision-making regarding the casting start time for continuous casters in steel production are core techniques for ensuring orderly and efficient production. The formulation of multi-cast batching plans is crucial for ensuring quasi-continuous and low-cost operation of continuous casting process. The execution of cast batching plans can be affected by the supply of molten iron. The fluctuation in blast furnace discharging rhythm and transportation time at the iron-steel interface has led to uncertainty in molten iron supply. Thus, this paper proposes a robust optimization method for formulation of multi-cast batching plans and dynamic decision-making on casting start times. This method not only optimizes the combining and sequencing of charge batching plans within multi-cast batching plans but also takes into account the conditions for continuous casting between different cast batching plans. It makes decisions on casting start times and adjusts adaptively considering the arrival of molten iron. An integer programming model M1 is established with the optimization objectives including the production efficiency, continuity degree of continuous casting plans, and comprehensive penalty values for deviations from the expected plan. Furthermore, considering the uncertainty in the arrival time of molten iron, a robust optimization model M2 for dynamic decision-making on casting start times is developed. A multi-objective hybrid-coding non-dominated sorting genetic algorithm is designed. This approach first obtains multi-cast batching plans, then dynamically calculates casting start time based on the online metal resource quantity, thereby obtaining real-time decision optimization schemes for cast batching plans and casting start time under dynamic molten iron arrival. The model was tested using manual decision-making data from the production performance of a specific steel plant. The results indicate that the model established based on the proposed method outperforms the classical NSGA-II algorithm and manual decision-making.