Hybrid Value Function Approximation for Solving the Technician Routing Problem with Stochastic Repair Requests.

This article focuses on the technician routing problem with stochastic requests (TRPSR), which involves jointly optimizing technician routes and spare part inventories for servicing geographically dispersed, uncertain repair tasks over multiple periods. The authors model TRPSR as a sequential decision problem and propose a novel hybrid solution combining genetic search with graph neural network (GNN)–based value function approximation, employing a unique multiattribute graph encoding with spatial markers to capture complex spatial-temporal dynamics without manual feature design. Extensive numerical experiments demonstrate that this hybrid approach outperforms benchmark policies in various realistic scenarios, effectively balancing travel, holding, fail-to-fix, and service delay costs while adapting to environmental changes. The study also provides managerial insights on spare part stocking and service prioritization, highlighting the importance of anticipatory spatial and temporal consolidation, especially under varying urgency and arrival rates of repair requests.