<b>Real-Time Multi-Depot Urban Logistics Optimization in Megacities via Transformer-Based Deep Reinforcement Learning</b>

This project is associated with a submission focused on "<b>Real-Time Multi-Depot Urban Logistics Optimization in Megacities via Transformer-Based Deep Reinforcement Learning</b>", which proposes two deep reinforcement learning models with Transformer architectures to optimize logistics distribution time costs across multiple depots in static and dynamic traffic scenarios, respectively.The "code" folder includes both the example data and code necessary to support the models for two logistics scenarios: DTM-MDVRP and DTM-DMDVRP. The "figure" folder contains the original images from the paper.The "Instruction" file offers detailed, step-by-step guidance for reproducing the research results.<b>Content</b><b>1.DTM-MDVRP</b><b>The "data" folder contains the</b> <b>example </b><b>data used by the </b><b>DTM-MDVRP </b><b>model and contains two files：</b>1)data/clientpoint.csv:This file contains logistics center and customer point data with fields including id and latitude and longitude;2)data/distance.csv:This file contains the travel time between nodes, as measured by road length and average travel speed;<b>The "Torch" folder contains the core code of the program:</b>1)baseline.py:This script implements the RolloutBaseline algorithm;2)config.py:This script is used to configure model training parameters；3)dataset.py:This script is used to generate the training dataset；4)train.py:Training scripts for models;5)validation_data_generator.py:This script generates the required validation data based on specified parameters;6)validation_test.py:This script tests the generated validation data.<b>The "Troch/Nets" folder contains the implementation of the model：</b>1)model.py:DTM-MDVRP model script；2)encoder_layers.py:This script includes the components of the encoder network architecture；3)encoder.py:This script contains the detailed implementation of the encoder；4)decoder_layers:This script includes the components of the decoder network architecture；5)decoder_utils:This script is used to implement the environment in deep reinforcement learning;6)decoder：This script contains the detailed implementation of the decoder.<b>The "ComparisonAlgorithms" folder contains the comparison algorithms selected for this study:</b>1)data_conversion.py: This script converts the input data into the required format, ensuring compatibility and consistency when running the comparison algorithms.2)route_visualization.py: This script visualizes the optimized delivery routes, supporting the analysis and comparison of the performance of different algorithms.3)ACO: The "ACO" folder contains the implementation of the Ant Colony Optimization (ACO) algorithm.4)SSA-SA:The "SSA-SA" folder contains the implementation of the SSA-SA algorithm, which combines the Sparrow Search Algorithm (SSA) and Simulated Annealing (SA) for route optimization.<b>2.</b><b>DTM-DMDVRP</b><b>The "data" folder contains the</b> <b>example </b><b>data used by the model and contains a folder and two files：</b>1)data/clientpoint.csv:This file contains logistics center and customer point data with fields including id and latitude and longitude;2)data/distance.csv:This file contains the travel time between nodes, as measured by road length and average travel speed;3)data/speed_data:This folder contains the dynamic travel time between nodes at different times, which is obtained by the length of the road and the real-time travel speed.The other directories and their functionality in DTM-DMDVRP are similar to those in DTM-MDVRP.<b>The environment required for this project includes:</b>Python 3.12; Pytorch; Scipy; Numpy.