EVPLAN: A GIS-enabled decision support tool for electric vehicle charging infrastructure planning

Abstract

The planning of electric vehicle charging infrastructure (EVCI) is a key challenge for cities undergoing energy transition. This paper introduces EVPLAN, a web-based application designed to support decision-makers in identifying optimal charging station locations through a flexible, multi-criteria approach integrated into an interactive geographic information system (GIS). EVPLAN combines technical and socio-economic indicators, with user-defined weights and thresholds. The results are visualized as heatmaps constructed from a hexagonal spatial grid, where each hexagon is assigned a suitability score calculated based on proximity to various infrastructure elements. A case study was conducted in the city of Marrakech, testing three planning scenarios. The outputs were then compared with a manually developed AHP (Analytic Hierarchy Process) model in ArcGIS, showing more than 80% overlap in high-priority zones, thus validating the reliability of the EVPLAN scoring method. In a second phase, EVPLAN's outputs were used as training data for several machine learning models (Random Forest, XGBoost, Neural Network, and TabTransformer) to assess their ability to learn and reproduce the spatial scoring logic. Random Forest achieved the highest accuracy (R 2 > 0.85), and its projection onto a different area of the city demonstrated strong spatial coherence. This hybrid approach shows that the structured outputs of a GIS-based decision tool like EVPLAN can be leveraged for predictive modeling, offering opportunities for territorial generalization and partial automation in EV charging infrastructure planning. • EVPLAN is a web-based interactive platform combining GIS and AHP for electric vehicle charging infrastructure planning. • The tool enables customized weighting of technical and socio-economic criteria with real-time visualization on a hexagonal spatial grid. • A case study in Marrakech showed over 80% overlap with results from a traditional AHP model implemented in ArcGIS. • EVPLAN outputs were used to train machine learning models (Random Forest, XGBoost, ANN, TabTransformer) achieving R 2 > 0.85. • The Random Forest model was successfully projected onto new urban areas, confirming its territorial generalization capabilities. • EVPLAN stands out as a flexible, accessible, and scalable decision-support tool aligned with sustainable electric mobility goals.