Author(s)

Wei Huang, Ruipeng Dong, Hongyu Liao

Corresponding Author

Wei Huang

Abstract

This paper aims to improve transportation efficiency and control costs by integrating LSTM, linear regression and integer programming to predict the volume of goods and dispatch vehicles in short-distance logistics transportation. First, the LSTM model optimized by Bayesian optimization is used to predict the daily volume of goods on each route, and then the prediction results are refined to 10-minute granularity through linear regression. In terms of scheduling strategy, the transportation task is divided into two stages: Stage 1 handles the whole volume of goods, and uses the greedy algorithm to prioritize the dispatch of owned vehicles to reduce costs; Stage 2 uses integer programming to generate a tail cargo point-to-point solution with the goal of minimizing total cost and maximizing the turnover rate of owned vehicles and the vehicle full load rate. In order to further improve efficiency, containerization operations are introduced to shorten the loading and unloading time to 10 minutes. Although the loading volume is reduced to 800 units, the model optimizes the turnover efficiency in stages by preferentially allocating containers to owned vehicles that can be dispatched again. Stage 1 calculates the vehicle return time after the container is used, and stage 2 uses the owned vehicles that return early to reduce the demand for external vehicles and outputs the decision on whether the vehicle uses the container. After introducing a random disturbance of about 5%, the results show that the turnover rate of owned vehicles remains stable, the vehicle loading balance fluctuates slightly, and the total cost is more sensitive to the prediction deviation, indicating that the model has a certain degree of robustness in terms of efficiency and load, but relies on prediction accuracy in cost control.

Keywords

LSTM Model, Linear Regression, Greedy Algorithm, Integer Programming, Terminal Transportation, Vehicle Scheduling