Data-Driven Optimization Research on Volume Prediction and Intelligent Scheduling in Logistics Sorting Centers

This study proposes a data-driven optimization framework to address the dual challenges of cargo volume prediction and intelligent scheduling in e-commerce logistics sorting centers. Leveraging historical operational data from JD Logistics, we develop an LSTM-based time series forecasting model enhanced with KNN for missing value imputation, achieving robust prediction accuracy across daily (average R²: 0.718) and hourly (average R²: 0.624) granularities. To optimize resource allocation, an improved PSO-BP neural network algorithm is introduced, incorporating chaotic Tent mapping and nonlinear inertia weight adjustments to balance global exploration and local exploitation, yielding a mean RMSE of 1248.7 for cargo flow predictions. A dynamic scheduling model integrating regular and temporary workforce constraints further minimizes operational costs while maintaining stability in transportation networks. Experimental results demonstrate the framework's effectiveness in handling volatility and fragmentation in logistics operations, offering a scalable solution for enhancing fulfillment efficiency and reducing costs through end-to-end prediction-planning integration.