A multi-stage stochastic-robust planning approach for highway service area self-contained energy system considering multiple uncertainties

The highway service area self-contained energy system (HSA-SCES) plays a crucial role in reducing global carbon emissions and advancing decarbonization goals. However, its practical deployment faces significant challenges, including the optimal configuration, the accommodation of multiple uncertainties, and the adaptation to dynamic load evolution. This paper proposes a multi-stage stochastic-robust planning approach for HSA-SCES to address these challenges. Firstly, the Bidirectional long short-term memory network based conditional generative adversarial network with an attention mechanism (BiLSTM-CGAN-Attention) model is proposed to construct representative multi-source uncertainty scenarios, and the improved k-medoids clustering method is employed for scenario reduction. Secondly, the traffic-flow-to-electric-vehicle (TF–EV) charging load mapping model is developed, integrating a dynamic EV penetration adjustment factor to faithfully characterize the temporal evolution EV charging demand. Finally, the multi-stage stochastic–robust bi-level optimization model is established and solved by integrating the Particle Swarm Optimization (PSO) algorithm with the Gurobi solver. The results demonstrate that the BiLSTM-CGAN-Attention model achieves superior performance in scenario generation, while the proposed planning approach outperforms deterministic and single-stage strategies, achieving cost reductions of at least 1.2 %.