Railway self-contained energy systems (RSCES) are crucial for sustainable rail development. However, there is a lack of optimization methods that account for dynamic grid interactions in self-consumption and surplus electricity injection scenarios. To address this, this paper proposes a topological architecture and configuration method for RSCES, employing a bilevel optimization approach combining an adaptive genetic algorithm with YALMIP/Gurobi. A case study at the Hailesihao South Traction Substation demonstrates that this model minimizes lifecycle costs despite higher initial investments, driven by electricity sales revenue and residual value. The analysis shows that while reducing transformer capacity can reduce initial capital costs, it leads to higher lifecycle costs due to reduced renewable energy feed-in capabilities, increased operational costs. Additionally, to avoid energy loss, the deployment scale of photovoltaic systems may need to be limited.