Anomaly Detection for HVAC System Maintenance Using Autoencoder Neural Network

The anomaly or outlier detection of HVAC system components enables the detection of system failures and unusual consumption patterns derived from system malfunctions. Prompt and effective anomalies detection of HVAC systems are imperative for initiating repairs, correct maintenance plans and eliminating errors in HVAC system energy consumption forecasts. In this paper publicly available datasets were used together with convolutional autoencoder neural network. In preparation of dataset for training neural network a procedure for preprocessing anomalous data is developed. The developed procedure and the neural network resulted in highly accurate detection of anomalies in HVAC system components.