Sustainable EV routing using spectral clustering and fuzzy reinforcement learning with energy constrained A* under mobility index and waiting constraints

Abstract

Abstract This research presents a comprehensive electric vehicle (EV) routing framework designed to address the complex interplay of real-world constraints in EV navigation. The proposed system integrates spectral clustering, fuzzy reinforcement learning, and enhanced pathfinding algorithms to compute optimal routes while considering battery limitations, traffic dynamics, terrain elevation, and charging station delays. Unlike conventional multi-objective EV routing solutions, which typically optimize metrics such as energy, time, and charging delays independently-this work addresses four major gaps in the field: (1) fragmented and isolated optimization lacking dynamic interdependency modeling, (2) limited real-time adaptability to traffic and charging dynamics, (3) inadequate topological modeling with respect to network clustering and geographic scalability, and (4) evaluation restricted to constrained environments. The system introduces four core innovations: (1) a topologically adaptive clustering mechanism using spectral clustering with geodesic distance metrics and elliptical regional modeling;(2) a time-dependent arrival simulation model that predicts charging station occupancy with high accuracy by incorporating temporal demand and station-specific dynamics; (3) a fuzzy reinforcement learning-based charging station evaluator that incorporates spatial density, occupancy trends, and temporal availability; and (4) an enhanced A* algorithm with integrated elevation-aware energy profiling, real-time traffic sensitivity, and adaptive SOC constraint modeling.Experimental evaluations conducted across diverse topographies demonstrate superior performance over baseline and established algorithms including Dijkstra, A*, Hybrid A*, and EVRP + Charging Aware techniques. The proposed method achieves a 22.8% reduction in total journey time (from 877 to 677 minutes), 19.6% improvement in energy efficiency (from 224.5 to 180.5 kWh), and a 63.3% decrease in waiting time (from 34.2 to 12.5 minutes) when compared to the traditional distance-based routing. Additionally, the system achieves a 90.0% reduction in battery violations (from 18.0% to 1.8%), addressing range anxiety through improved SOC-aware planning.The findings confirm that the framework advances beyond both established algorithms and recent multi-objective solutions, offering a more unified and effective approach to EV routing. Performance gains remain consistent across urban, rural, and elevation-intensive routes, with measured improvements of up to 27.2% over conventional routing algorithms. This research directly contributes to SDG 7 (Affordable and Clean Energy) and SDG 11 (Sustainable Cities and Communities) by enabling energy-efficient, reliable EV navigation, thereby supporting the broader vision of clean transportation and smart city integration.