Research on Freight Volume Variation in Logistics Networks Based on the Combined Model of SSA-LightGBM-BiLSTM

This paper presents a study on the variation of freight volume in logistics networks using a combined model of SSA-LightGBM-BiLSTM. Initially, the logistics network’s topological structure and exogenous variables are constructed. Subsequently, the route impact index is computed based on the exogenous variables, thereby deriving the index values for each sorting center. These index values quantify the expected impact of route changes on the total freight volume for each sorting center. Considering the lack of clear seasonal and cyclical features in the data, the paper introduces the Bi-LSTM and LightGBM models. The SSA algorithm is employed to optimize model hyperparameters, leading to the establishment of the SSA-LightGBM-BiLSTM combined model based on exogenous variables. Each model undergoes 10-fold cross-validation to accurately capture the freight volume fluctuations caused by route changes. Finally, in conjunction with VMD denoising technology, the paper achieves predictions of daily and hourly freight volumes for 57 sorting centers over the next 30 days based on transportation route changes. The combination of multiple models’ predictions enhances overall prediction accuracy and robustness.