Stochastic MILP Model for Merging EV Charging Stations with Active Distribution System Expansion Planning by considering Uncertainties

Abstract

Radial Power Distribution Networks (PDNs) often suffer from limited reliability, flexibility, and efficiency, leading to service interruptions. Planning for radial PDNs is essential to enhance redundancy resilience, reduce disruptions, and improve overall efficiency. However, traditional PDN planning methods have become obsolete due to the proliferation of Distributed Generation (DG) resources and energy storage systems. Additionally, the rise of Electric Vehicles (EVs) demands sophisticated charging infrastructure planning. This article presents a Mixed-Integer Linear Programming (MILP) model for joint expansion planning of PDN and Electric Vehicle Charging Stations (EVCSs). The model takes into account the construction or reinforcement of substations and circuits, along with the integration of EVs, the installation of DGs, and the placement of capacitor banks, all regarded as traditional conventional expansion options alternatives. To address uncertainties associated with DG generation, conventional loads, and EV demand, our model identifies optimal installation and asset locations. We formulate this as a stochastic scenario-based program with chance constraints for Power Distribution Network Expansion Planning (PDNEP), minimizing investment, operational, and energy loss cost costs over a planning horizon. Through two deterministic and stochastic approaches, encompassing six case studies on an 18-node test system, we evaluate the effectiveness of our model. Results are further validated on a 54-node system, confirming the model’s robustness. Notably, the numerical findings underscore the substantial cost reduction achieved by including EVCSs in the stochastic expansion planning approach, demonstrating its cost-effectiveness. In case study I, where all EVs charge at home during peak hours, it’s the worst case for the PDN. The 54-node system, more complex, demands longer computational time. In the 18-node system, costs improve from 9.97% (case study II) to 3.96% (case study VI) versus the worst-case (case I). In the 54-node system, improvements range from 10.47% (case study II) to 1.40% (case study VI). As a result, In comparative analyses against deterministic and stochastic approaches, our model consistently outperforms in diverse test case studies. The proposed model’s adaptability to address uncertainties underscores its suitability for solving the PDNEP problem in PND.