Data-driven modelsfor estimating heatpump powerconsumption

Abstract: The number of installed heat pumps has been rapidly increasing in recent years, acceleratingthe decarbonisation of the heating sector. The impact of the increasing deployment of heatpumps on the grid can be evaluated with the help of models estimating the heat pumppower consumption. This thesis contributes to the field of data-driven heat pump modellingby developing regression models based on data from field installations to reflect the heatpump operation in real conditions. The developed models estimate the heat pump powerconsumption using a limited number of input features (parameters) measured during heatpump operation. This thesis analysed anonymised data obtained from the monitoring systemof domestic ground-source heat pump (GSHP) and air-source heat pump (ASHP) installationsto develop GSHP and ASHP regression models. Prior to developing the regression models, the data were pre-processed and the most important features (measured parameters) used asindependent variables in the regression models were identified. Further, various regressionmodels were proposed ranging from simple-linear, multiple-linear to non-linear (up to thefourth-degree) regression models, with and without the interaction terms and with the varyingnumber of the selected input features. The identified most significant input features fordeveloping the regression models based on the obtained datasets involved supply, sourceand outdoor temperatures and compressor frequency. The results in this thesis showed thatregression models can estimate the heat pump power consumption with a satisfactory accuracy(up toR2 88 % and mean absolute percentage error 13 %). Furthermore, it was proven that nonlinearregression models performed with higher accuracy compared to linear regression modelsand the accuracy was increasing with the increasing number of statistically significant inputfeatures. This thesis also revealed the importance of outlier detection and feature selectionprior to developing heat pump models when data from field installations are used.