OPSEARCH Management Information Systems
Adaptive dual-feedback multi-objective ACO framework for dynamic electric vehicle routing under energy constraints
Abstract
This study develops an adaptive dual-feedback multi-objective ant colony optimization (ACO) framework for the Electric dynamic vehicle routing problem. The method maintains separate pheromone matrices for distance and energy and combines them during route construction through stratified $${\uplambda}$$ -sampling, ensuring representation of both boundary objectives and intermediate trade-off regions in every iteration. A non-dominated archive provides additional objective-specific pheromone reinforcement, while archive spread is used to adapt the distance and energy evaporation rates. EV-aware construction incorporates escape-energy screening, charging-station repair, and $$2$$ -opt local improvement to preserve battery feasibility. The proposed method was evaluated on a $$200$$ -customer instance over $$30$$ paired runs and achieved a mean composite cost of $$2020.5522$$ compared with $$2038.9019$$ for controlled standard ACO, corresponding to a baseline-normalized improvement of $$0.900\%$$ . The paired difference was statistically significant (Wilcoxon signed-rank test, $$p=0.00873$$ ), with the proposed method obtaining the lower cost in $$22$$ of the 30 paired runs. Additional rolling-horizon experiments showed $$100\%$$ request acceptance in the tested dynamic scenarios, while tests on six Solomon-derived instances indicated consistent performance across clustered, random, and mixed customer distributions. Expanded ablation analysis showed that GA crossover and mutation do not provide a measurable benefit under the tested conditions; therefore, these operators and their AET-based trigger are not included in the final proposed method.