Abstract
Vapor injection is a relative new technique used on air source heat pumps to enhance the coefficient of performance under low outdoor temperatures. The new system – vapor-injection air source heat pump (VIASHP) – has received rather limited research in the past, especially in numerical studies. Therefore, this paper presents a Bayesian regularized neural network model for predicting the energy performance of VIASHPs operating in residential buildings. The model is trained and validated by on-site measured data which is collected in 1-minute interval from two random selected residential buildings over 4 months. A global sensitivity analysis of the measured data is conducted to identify the more important parameters as network inputs. We also process the measured data into 5-minute, 10-minute, 30-minute, 1-hour, 2-hour, 1-day and 1.5-day intervals to study the impacts of temporal resolution, time delay and number of hidden neurons on model prediction accuracy. The results show good agreement between measured data and neural network modeled data. Following findings are obtained in the two residential buildings: (1) lower temporal resolution and time delay can improve the prediction performance. However, (2) with the decrease of temporal resolution, the positive impact of time delay is declining. (3) when the temporal resolution is daily, best prediction performance can be achieved when no time delay is implemented. (4) Increasing the number of hidden neurons can improve the prediction performance for high temporal resolution, but deteriorate it for low temporal resolution.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have