An Interpretable Socio-Technical Decision Support System for Bi-Objective Urban Distribution Center Location: Adaptive Optimization Supervised by a Large Language Model

Urban e-commerce fulfillment involves multiple operational stages, including goods receipt, storage, picking, packaging, dispatching, and delivery to customers. This study focuses on one strategic component of this broader fulfillment process: the location–allocation design of urban distribution centers. We develop a socio-technical decision support system for bi-objective urban distribution center planning, in which ex ante location decisions determine which candidate facilities should be opened, whereas ex post allocation decisions assign demand points to the selected facilities under service-time constraints. The model jointly minimizes total logistics cost and population-weighted delivery time, seeking a synergistic balance between cost efficiency and service responsiveness rather than optimizing either objective in isolation. We further embed a large language model into NSGA-II as a bounded supervisory controller that periodically diagnoses search states, adjusts operators and probabilities, and records structured adaptation logs. Experiments on the ZDT benchmark suite and a real urban case demonstrate that this approach improves optimization performance while producing reviewable intervention records. The study contributes to systems research by organizing adaptive AI supervision and organizational oversight into an integrated urban logistics planning system.