Dynamic route optimization in smart logistics using hybrid machine learning models

Purpose This study aims to address dynamic route optimization in smart logistics by tackling persistent inefficiencies and sustainability challenges. It proposes a scalable, adaptive hybrid model designed to enhance operational efficiency and promote environmental responsibility in intelligent transportation systems. Design/methodology/approach A hybrid model integrates Transformers for temporal modeling, graph neural networks (GNNs) for spatial relationships and generative adversarial networks (GANs) for diverse route generation. Experiments use MIT Reality Mining, UTM and proprietary data sets, implemented in Python 3.9, PyTorch 1.12.1, PyTorch Geometric 2.0.4 and Hugging Face Transformers 4.20.0. Findings The model achieves an average of 15.3% travel distance reduction, 20.7% time efficiency improvement and 10.2% energy consumption decrease across data sets, outperforming baseline methods. Originality/value The novel Transformer-GNN-GAN framework uniquely integrates temporal, spatial and generative modeling, advancing sustainable logistics and offering insights for multimodal transport and autonomous systems.