A GeoAI-Enabled Spatial Capacitated Allocation Network for Waste Collection Facility Siting with Exclusion Zones

Determining locations for waste collection points that are practicable and efficient is a core challenge for municipal solid waste management. Planners should simultaneously balance coverage, walking distance, capacity limits, and environmental siting constraints such as buffers around sensitive areas. Existing approaches tend to address only parts of this problem: Geographic Information System–based multi-criteria decision analysis (GIS–MCDA) studies focus on suitability mapping without explicit capacity or assignment; classical facility location models assume idealized settings and rigid hard constraints with limited exclusion handling; and clustering methods emphasize geometric compactness but rarely incorporate exclusion zones or operational capacity ranges. In rural and small-town settings, these gaps are often bridged by manual siting rather than through reproducible, data-driven tools. This paper proposes the Spatial Capacitated Allocation Network (SCAN) model, a capacity-constrained spatial clustering framework for siting waste collection points in both dense urban neighborhoods and dispersed rural townships. SCAN combines a mixed-integer assignment step with continuous updates of facility locations in an Alternating Assignment–Relocation (AAR) algorithm, a coordinate-descent procedure inspired by k-means, extended to handle capacity bands, soft service-radius penalties, and exclusion zones within a unified workflow. Experiments on a compact high-rise neighborhood in Shenzhen and a mountainous rural county in Sichuan, China, show that SCAN attains high coverage, short access distances, and good capacity utilization while respecting GIS-derived exclusion constraints where planning often relies on manual site selection. Together, these results demonstrate that SCAN provides a flexible decision-support tool for geospatially aware planning of waste collection infrastructure across settlement patterns.