To charge or to transfer? Coupled scheduling in smart operating rooms with reinforcement learning-based neighborhood search

Powered patient-transfer beds and smart transport systems are increasingly being introduced in hospitals to reduce the manual-handling burden and support patient movement. When such equipment is used for time-critical transfers, finite battery capacity and state-of-charge-dependent availability can couple transport decisions with operating-room schedules. This paper formulates a heterogeneous three-stage surgical scheduling problem that jointly optimizes patient selection, clinical-stage sequencing, PHU/OR/PACU resource assignments, ITB transport-task assignment and order, and SOC-triggered charging. Unlike conventional surgery-scheduling models that treat inter-stage transport as exogenous or resource-free, and unlike AGV logistics models that take clinical service processes as given, the proposed formulation makes clinical and transport-energy decisions mutually endogenous: the clinical schedule releases transport tasks, whereas ITB availability and charging feed back into patient arrival times and medical-resource utilization. The resulting problem has a large decision space with tightly coupled scheduling and energy constraints. We propose QH-MLNS, a reinforcement-learning-enhanced, multi-destruction-strength large neighborhood search algorithm that combines a destroy–repair strategy with multiple destruction strengths, Q-learning-based adaptive operator selection, and dynamic dual-pool retention. The study examines a deployment regime in which battery-powered transfer devices are shared and charging can restrict their availability; it does not assume that such devices are already standard in all hospitals. On the tested resource-constrained and resource-abundant scenarios, QH-MLNS yields 30.3% and 52.5% higher mean hypervolume and 45.6% and 96.2% lower mean inverted generational distance than H-MLNS and MLNS, respectively.