Research on short-distance transportation prediction and scheduling method based on multimodel fusion and intelligent optimization

As an important link at the end of the logistics system, short-distance transportation is characterized by strong timeliness, high pressure on resource utilization and complex scheduling. Aiming at the shortcomings of traditional scheduling methods in terms of prediction accuracy and resource allocation efficiency, this paper proposes a short-haul transportation prediction and scheduling research framework that integrates multimodel learning and intelligent optimization. Firstly, a Stacking integrated prediction model based on LSTM, Light GBM and SVR is constructed to predict the parcel cargo volume of each route with high accuracy, and a 10-minute granularity time splitting is realized by combining the historical distribution pattern and local regression method. Secondly, for the scheduling optimization problem, a multi-stage hybrid scheduling model containing greedy inspiration, local search and taboo search (Tabu Search) is designed to improve the turnover rate of owned vehicles and overall transportation efficiency. In terms of container use strategy, this paper introduces the XG Boost classifier to construct an intelligent container discrimination model and explains its decision-making process through the SHAP method, so as to realize the balance between loading and unloading efficiency and loading capacity. Finally, a prediction deviation sensitivity analysis system is constructed to comprehensively evaluate the impact of various types of perturbations on the scheduling performance and verify the robustness of the model under systematic deviation and peak perturbation scenarios. The experimental results show that the method proposed in this paper is better than the traditional method in terms of prediction accuracy, scheduling performance and adaptability, and is suitable for complex and changing end-to-end logistics scenarios.