Fusion-modeling-driven adversarial coevolutionary robust MOPSO for the zinc leaching process under feed disturbances

Operating set-point optimization in zinc leaching processes (ZLPs) is essential for resource utilization, process stability, and impurity control. Adjusting waste acid and zinc calcine feeds regulates pH, reaction intensity, zinc dissolution, and ferric ion removal across a continuous stirred tank reactor (CSTR) system comprising five tanks. However, complex reaction mechanisms, uncertain feed disturbances, and coupled objectives and constraints make robust optimization difficult. This paper proposes a fusion-modeling-driven adversarial coevolutionary robust multiobjective particle swarm optimization algorithm (FD-ACR-MOPSO) to optimize the operating set-points of multiple material flows. First, a fusion-driven dynamic prediction model combines CSTR mass balances with XGBoost kinetic compensation to improve response evaluation under varying inlet conditions. Second, operation and disturbance populations coevolve to identify adverse scenarios and apply disturbance pressure according to the search roles of operating solutions. Third, process risk feedback adapts particle learning and archive guidance, steering the swarm away from sensitive boundaries and toward robust feasible regions. Comparative and ablation experiments show that FD-ACR-MOPSO achieves a favorable balance among robust zinc production, unit operating cost, and outlet ferric ion control under the tested disturbances.