China’s Dual-Carbon Policy: A Two-Stage Hybrid Assessment Framework for Provincial Crude Steel Capacity-Adjustment Pressure Using XGBoost and SHAP

Against the backdrop of China’s dual-carbon goals—peaking carbon emissions by 2030 and achieving carbon neutrality by 2060—capacity optimization in the steel industry can no longer rely solely on aggregate output reduction. Instead, governance must shift toward a multidimensional approach that combines scale control, structural adjustment, and coordinated regional allocation. This study develops a quantifiable and interpretable assessment model for capacity-adjustment pressure. Monthly provincial crude steel output is used as a high-frequency proxy for capacity utilization and production adjustment. Additive time-series decomposition is applied to extract three components from monthly output—trend, residual, and volatility—which respectively represent structural evolution, short-term deviations, and exposure to shocks. The model further incorporates multidimensional variables, including downstream steel demand, resource and transport constraints, scrap steel ratio, and policy constraints. On this basis, a two-stage hybrid assessment framework is developed. In the first stage, extreme gradient boosting (XGBoost) is used to learn nonlinear relationships and derive data-driven feature importance. In the second stage, a composite pressure index is constructed and transformed into a standardized 0–100 score through a robust rank-based mapping mechanism. Dual thresholds are then used to generate three policy recommendations: maintaining current capacity, structural optimization, and capacity reduction. The results show that production trends and volatility intensity are the primary drivers of capacity-adjustment pressure, while pronounced spatial heterogeneity requires highly localized strategies. The classification assigns 25 provinces to maintaining current capacity, 3 to structural optimization, and 3 to targeted capacity reduction. Finally, integrating SHapley Additive exPlanations (SHAP) enhances model interpretability and provides a quantitative basis for shifting from indiscriminate capacity suppression toward differentiated, region-specific capacity governance, thereby supporting the sustainable low-carbon development of the global steel industry.