Distributionally robust bi-level optimization for coal-dominant generation expansion planning under deep uncertainty: A decision-dependent framework for Bangladesh’s energy transition

This paper develops a distributionally robust bi-level optimization (DR-BiLevel) framework for generation expansion planning that explicitly models decision-dependent uncertainty through endogenous Wasserstein ambiguity sets. The framework captures how renewable investments increase operational uncertainty while storage investments reduce it—a dynamic ignored by conventional methods. Applied to Bangladesh’s power sector (28.9 GW installed capacity), the DR-BiLevel approach yields expansion plans 8.5% less costly than standard distributionally robust optimization and 20.8% less costly than robust optimization, while maintaining 95% reliability guarantees. The enhanced column-and-constraint generation algorithm achieves finite convergence with practical computation times (8.5 h for the full case study). Results suggest a pragmatic transition pathway: complete committed coal projects (8.2 GW) but halt new coal development; achieve 35% renewable penetration by 2041 through 18 GW solar and 5 GW offshore wind deployment; deploy 4.8 GW nuclear capacity; and expand storage to 4 GW. Carbon pricing sensitivity analysis identifies $25–35/tCO2 as the economic threshold triggering rational fuel switching. The framework demonstrates that explicitly modeling decision-dependent uncertainty fundamentally changes optimal strategies, providing actionable guidance for energy transition planning in coal-dominant developing economies.