A fuzzy multi-agent framework for hospital capacity assessment based on a dedicated resource agent

This is the authors' abstract. We don't add key points for this paper.

Hospital operational capacity cannot be reliably represented by a single occupancy indicator because it emerges from the interaction of heterogeneous and interdependent resources, including staffed beds, medical equipment, consumables, sterilization processes, and operational delays. This study introduces a dedicated fuzzy Resource Agent designed to continuously assess the internal operational state of hospital resources independently of patient-demand dynamics. By separating resource-capacity assessment from scheduling and demand management, the proposed agent provides an interpretable and reusable capacity representation that can support higher-level coordination within hospital multi-agent systems. The proposed Resource Agent integrates 17 normalized resource indicators through a 47-rule weighted Mamdani inference system complemented by an explicit critical-resource veto that prevents severe shortages from being masked by favorable secondary resources. The framework was evaluated using a reproducible in silico experimental protocol comprising 12 hourly scenarios observed over 720 h: one normal control and 11 degradation, recovery, oscillatory, abrupt, or isolation scenarios. Threshold calibration used 200 independently generated normal-condition trajectories, and comparative evaluation used 1,000 paired Monte Carlo replications. The restricted time-to-alert estimand assigned non-detections the remaining degraded follow-up. Mean restricted time-to-alert was 3.953 h for the Resource Agent and 73.406 h for the deterministic comparator; excluding the influential near-threshold scenario reduced the exploratory mean gap from 69.453 h to approximately 32.46 h. A literature-informed Fuzzy TOPSIS adaptation yielded 38.102 h overall, whereas the median scenario-level TOPSIS-minus-Resource Agent difference was 0.000 h, indicating comparable performance across most scenarios while highlighting the specific contribution of the critical-resource veto during isolated critical failures. No veto-ceiling violations occurred in 10,000 randomized trials, and only four material quasi-monotonicity inversions were observed in 1,000,000 trials. These findings demonstrate the computational behavior and architectural value of the proposed Resource Agent under a reproducible experimental protocol and provide a computational foundation for future validation using real hospital operational data.