Paper Conference

Proceedings of eSim 2020: 11th Conference of IBPSA-Canada


Analysis of Feature Importance in Modeling Ground Source Heat Pump Systems Using Broad Parametric Analysis, Load Characterization and Artificial Neural Networks

David Rulff 1,2, Kevin Cant 1, Theodor Victor Christiaanse 1, Ralph Evins 1
1 University of Victoria, Canada

Abstract: This paper considers the case of modeling a ground source heat pump with a range of temporal load dynamics to identify the important features used for estimating performance. Heating and cooling load profiles are generated using extensive parametric sampling of a base office building simulation, including variation of a set of parameters for heat pump system design and properties of the ground. Load characteristics are extracted from the models using aggregate output and application of Fourier Transform decomposition to describe periodic behaviour. Artificial neural networks are used to estimate the heating and cooling performance metrics of the ground source heat pump system, with significant accuracy using the full feature set (R²>0.98). The resulting loss in accuracy due to reduced dimensionality through feature grouping is also shown, with implications for early stage design and performance modeling.