Fairness-Oriented Charging Station Location Optimization Driven by Deep Reinforcement Learning

Aim:

            The aim of this project is to develop a fairness-oriented EV charging station location optimization framework using geospatial analytics and deep reinforcement learning. It integrates population density, POI distribution, and spatial coverage to identify high-impact candidate sites. The system ensures region-balanced accessibility while maximizing demand-weighted service coverage.

Abstract:

            This project presents a GIS-driven EV charging station planning framework enhanced by a deep reinforcement learning model inspired by SpoNet. Population raster data, POI density, and spatial grid structures are combined to estimate localized charging demand and generate candidate locations. A modified pointer-network-based RL algorithm optimizes station placement by jointly maximizing coverage and regional fairness. Redundancy pruning and demand-weighted coverage evaluation further refine the final solution for practical deployment. The system delivers a data-driven, equitable, and operationally viable approach for EV charging infrastructure expansion in urban environments.

Proposed System:

        The proposed system extends the SpoNet framework by integrating full GIS-based preprocessing, including raster population extraction, POI aggregation, and spatial grid generation. Instead of using coarse administrative districts, K-means clustering creates refined pseudo-regions to enable more granular fairness assessment. Spatial coverage is computed through real geospatial buffers and intersection operations, providing accurate service modeling across the city. A simplified, practical DRL pointer network selects candidate sites based on population, POI density, and regional balance. A region-deficit weighting mechanism guides the selection toward underserved zones, improving fairness beyond the original formulation. Redundant stations are automatically pruned using coverage overlap analysis, ensuring efficient infrastructure deployment. Demand-weighted coverage evaluation combines population and POI intensity to better reflect real-world charging needs. Interactive Folium mapping visualizes coverage, station density, and service areas, making the system directly usable for urban planning.

Advantage:

• The system transforms the theoretical SpoNet model into a fully operational GIS-enabled solution suitable for real-world deployment.
• It provides more accurate spatial coverage using true geographic buffers rather than symbolic coverage matrices.
• Pruning redundant stations and weighting demand by POIs significantly improves efficiency and reduces unnecessary infrastructure cost.
• The addition of dynamic visualization and spatial diagnostics allows planners to interpret, validate, and adjust station placement with clarity.