Comparative Analysis of Methods for Predicting Brine Temperature in Vertical Ground Heat Exchanger—A Case Study

• Published in: Energies (Basel) Vol. 17; no. 6; p. 1465
• Main Authors: Piotrowska-Woroniak, Joanna, Nęcka, Krzysztof, Szul, Tomasz, Lis, Stanisław
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.03.2024
• Subjects: brine temperature from vertical ground heat exchangers; brine water heat pumps; CART; Case studies; CHAID; Climate; Collaboration; Earth temperature; Efficiency; Energy consumption; Equipment and supplies; Force and energy; Geothermal power; Heat conductivity; Heat exchangers; Heating; heating system energy efficiency; Methods; Phase transitions; prediction model based: ANN; Radiation; Seasons; Set theory; Temperature; Thermal energy; Variables; Weather forecasting; Poland
• ISSN: 1996-1073; 1996-1073
• DOI: 10.3390/en17061465

This research was carried out to compare selected forecasting methods, such as the following: Artificial Neural Networks (ANNs), Classification and Regression Trees (CARTs), Chi-squared Automatic Interaction Detector (CHAID), Fuzzy Logic Toolbox (FUZZY), Multivariant Adaptive Regression Splines (MARSs), Regression Trees (RTs), Rough Set Theory (RST), and Support Regression Trees (SRTs), in the context of determining the temperature of brine from vertical ground heat exchangers used by a heat pump heating system. The subject of the analysis was a public building located in Poland, in a temperate continental climate zone. The results of this study indicate that the models based on Rough Set Theory (RST) and Artificial Neural Networks (ANNs) achieved the highest accuracy in predicting brine temperature, with the choice of the preferred method depending on the input variables used for modeling. Using three independent variables (mean outdoor air temperature, month of the heating season, mean solar irradiance), Rough Set Theory (RST) was one of the best models, for which the evaluation rates were as follows: CV RMSE 21.6%, MAE 0.3 °C, MAPE 14.3%, MBE 3.1%, and R2 0.96. By including an additional variable (brine flow rate), Artificial Neural Networks (ANNs) achieved the most accurate predictions. They had the following evaluation rates: CV RMSE 4.6%, MAE 0.05 °C, MAPE 1.7%, MBE 0.4%, and R2 0.99.