Patient visits-based comprehensive energy consumption prediction for hospital in China: A machine learning approach take hot and humid region as example

• Published in: Journal of Building Engineering Vol. 111; p. 113593
• Main Authors: Li, Lingling, Li, Yuan, He, Jingyuan, Su, Ruibo, Shi, Yuan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.10.2025
• Subjects: Energy consumption; Feature selection; Hospital building; Machine learning; Patient visits; Patient visits; Energy consumption; Feature selection; Hospital building; Machine learning
• ISSN: 2352-7102; 2352-7102
• DOI: 10.1016/j.jobe.2025.113593

Hospital buildings, serving as restorative environments, exhibit significantly different energy demands and operational patterns compared to typical public buildings. The energy consumption of hospital building is highly correlated with patient visits. However, influenced by factors such as doctor scheduling, weather conditions, and unexpected events, patient numbers demonstrate highly dynamic and irregular variation patterns. This makes it difficult for traditional prediction models that rely solely on historical energy consumption patterns to effectively solve such complex problems, and the prediction accuracy is limited. Therefore, developing new prediction models that can integrate multi-source information is crucial for improving the accuracy and applicability of energy consumption forecasts in hospital outpatient buildings. To enhance prediction precision and applicability, this study proposes machine learning models integrating hospital-specific functional features (patient visits) and general features (climate, time factors). Utilizing 2023 hourly energy, patient visits, and climate data (8760 data sets) from a representative general hospital in a hot-humid region, the research employed Spearman correlation analysis to screen key features and evaluated six algorithms across diverse input scenarios. Results indicate: 1) Total outpatient visits strongly correlate (coefficients >0.50) with total, lighting, elevator, and diagnostic equipment energy consumption; 2) Deep learning model of One-Dimensional Convolutional Neural Network achieved better performance, with R-values of 0.92, the mean absolute percentage error of 7.83 %. The study identifies key influencing parameters and suitable algorithms for hospital energy consumption prediction. The proposed models incorporating outpatient visits improve accuracy and reduces training time, helping hospitals optimize energy management. This approach is particularly beneficial for general hospitals with high outpatient volumes. •Analyze the energy consumption structure of a hospital outpatient building.•Identify patient visits as a key feature for hospital energy consumption prediction.•Integrate multi-source data to optimize prediction model.•Evaluate the performance of six models in three input feature scenarios.