Development of Anomaly Detectors for HVAC Systems Using Machine Learning

Faults and anomalous behavior affect the operation of Heating, Ventilation Air Conditioning (HVAC) systems. This causes performance loss, energy waste, noncompliance with regulations discomfort among occupants. To prevent damage, automated, fast identification faults in HVAC systems is needed. Fault Detection Diagnosis (FDD) techniques are very effective for these purposes. The best FDD methods, terms cost effectiveness data exploitation, based on process history; i.e., sensor from automation In this work, supervised semi-supervised models were developed. Other than regard to outdoor temperature humidity, input parameters an system have few internal variables. Performance traditional methods (e.g., VAR, Random Forest) low, so Artificial Neural Networks (ANNs) selected, since they can capture nonlinear relationships features easily optimized. ANNs detect simultaneous different classes. ANN metrics evaluated. ground truth obtained history (supervised case) a mix deterministic clustering (semi-supervised case). derivation case, extensive comparison advanced models, set work apart previous studies. Mean Absolute Error (MAE) model was 0.032 over 15 min 0.034 30 min. Balanced Accuracy Score (BAS) 86%.