Comparative analysis of metaheuristic techniques for logistics 4.0 optimization: Dynamic vehicle routing and LSTM hyperparameter tuning

Logistics 4.0 integrates IoT, artificial intelligence, and real-time data streams, creating dynamic and high-dimensional optimization challenges across supply chain operations. This study provides a rigorous theoretical and empirical comparison of four metaheuristic algorithms—Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Simulated Annealing (SA)—applied to two interconnected problems: the Dynamic Vehicle Routing Problem (DVRP) and hyperparameter optimization of LSTM models for demand forecasting. Using standardized benchmarks (dynamic TSPLIB instances and a stratified Rossmann Store Sales dataset), algorithms are evaluated through a Composite Performance Index (CPI) that integrates solution quality, computational time, scalability, and statistical robustness. Results show that PSO’s velocity-driven continuous search achieves threefold faster convergence in LSTM tuning than modern Bayesian methods under fixed 1000-evaluation budgets, while ACO’s pheromone-based constructive mechanism produces 4–8% superior routing solutions within 300 s latency constraints. The DVRP is further extended to a bi-objective sustainability framework, where ACO and GA reduce CO 2 emissions by 11–12% with an approximately 4% increase in travel distance via strategic weight-shedding. Finally, a Phase-Aware Adaptive Selector (PAAS) is introduced to coordinate algorithm deployment across operational phases, achieving performance comparable to an oracle selector without additional selection overhead. The findings establish a unified empirical baseline linking algorithmic search dynamics to operational constraints in sustainable Logistics 4.0 systems.