Data-driven multi-objective operational optimization and decision support for copper side-blown smelting under varying conditions

Operational optimization of copper side-blown smelting is challenging because feed composition and furnace conditions vary over time, while multiple conflicting metallurgical targets must be coordinated simultaneously. This study develops a data-driven operational decision-support framework for repeated multi-objective optimization under varying operating conditions. A heterogeneous ensemble surrogate combining support vector regression (SVR), LSBoost, and random forest (RF) with domain-informed features is constructed from 7298 industrial production samples and achieves an average R 2 of 0.8466 on 1460 held-out samples for four key metallurgical indicators. To exploit continuity between successive operating conditions, an Elite Memory Injection SPEA2 strategy (EMI-SPEA2) reuses historical Pareto solutions during re-optimization. Specifically, high-quality Pareto solutions accumulated under previous operating conditions are injected into the initial population to warm-start the evolutionary search under the current condition, yielding a 77.4% higher mean hypervolume than the strongest benchmark algorithm. Across 19 chronologically sampled operating instances, EMI-SPEA2 achieves a mean inverted generational distance (IGD) of 0.1416 and hypervolume (HV) of 0.8624, with statistically significant HV improvements over six benchmark algorithms. A combined subjective–objective weighting technique for order preference by similarity to ideal solution (TOPSIS) procedure subsequently converts each Pareto set into a coordinated recommendation for five manipulated variables while satisfying prescribed metallurgical constraints. The results demonstrate the potential of historical-knowledge-assisted multi-objective optimization for operator decision support under changing smelting conditions.