Framework for Hospital Operational Anomaly Detection using Prediction Error Analysis of an XGBoost-based Cooling Load Model

초록

The energy consumption patterns of hospitals, as critical infrastructure, have complex characteristics directly linked to operational stability. This study moves beyond simple prediction to propose a novel framework for real-time operational anomaly detection in hospitals by leveraging the prediction errors of a forecasting model. To achieve this, a high-fidelity baseline model that simulates the highly regular ‘normal behavior' of a hospital's cooling load was first constructed using the XGBoost algorithm. The proposed framework identifies events that deviate from normal operations by defining an ‘Anomaly Score' from the prediction error, which is calculated as the difference between the model's prediction and the actual value. Time-series crossvalidation results demonstrated that the normal behavior model achieved very high prediction accuracy (with a coefficient of determination over 0.89), proving its reliability as a baseline. Furthermore, a simulation experiment with an injected synthetic anomaly empirically validated that the Anomaly Score spiked sharply during the event, confirming its capability to successfully detect anomalies. The key contribution of this work lies in reinterpreting a conventional prediction model as a real-time monitoring tool for the operational stability of critical facilities. This approach provides a practical foundation for shifting the energy management paradigm of hospitals from reactive response to proactive and preemptive management by enabling the early detection of potential equipment failures or operational inefficiencies.

키워드