Data Driven Model for Performance Evaluation and Anomaly Detection in Integrated Air Source Heat Pump Operation

In this paper we focus on a practical application of data driven process monitoring for performance evaluation and anomaly detection in the operation of air source heat pumps (ASHPs). At our test site, which is a hospital building in Singapore, we install sensors to monitor the operation of a hot water heating system (10 years old) comprising of 5 ASHPs operating according to a pre-defined switching algorithm to balance their workload. We collect data for 6 consecutive months. With the help of the data and site specific conditions, we first develop a Coefficient of Performance (CoP) based technique to visualize performance degradation regarding to ASHPs and to identify patterns for early fault detection. In our monitored data, three pumps malfunctioned with easily identifiable patterns, such as, consistent under-performance or over-performance followed by longer periods of under-performance. Then, using one month data, we also train a linear regression model for each ASHP performance. This model can also be used to predict the outlet temperature as well as the performance degradation in the operation of individual ASHPs relative to the performance in the training period. The results of the linear model are consistent with our CoP technique. The proposed data driven methods can yield huge economic benefits for the hot water system operator.