APPLICATION OF AN ARTIFICIAL NEURAL NETWORK FOR PREDICTING THE THERMAL STATE OF AN AIR-SOIL HEAT EXCHANGER

• Datum: 2025
• Hauptverfasser: Basok , B., Novitska, M., Zurian , O., Sylniahin , M., Davydenko , D.
• Format: Artikel
• Sprache: Ukrainian
• Veröffentlicht: Institute of Renewable Energy National Academy of Sciences of Ukraine 2025
• Schlagworte: artificial neural network; air-soil heat exchanger; modeling; thermal regime prediction; thermal state of the soil.
• Online Zugang: https://ve.org.ua/index.php/journal/article/view/537
• Назва журналу: Vidnovluvana energetika

In this paper, an artificial neural network was used to predict the thermal state of an air-soil heat exchanger. The artificial neural network was trained, tested, and validated using experimental data obtained at the Institutes of Engineering Thermophysics and Renewable Energy of the National Academy of Sciences of Ukraine. A simple Sequential neural network was used for calculations. The input parameters for the neural network were the air temperature at the heat exchanger inlet, its humidity, and the outlet temperature for the previous hour. The Tensorflow API and its Keras module were used for calculations. The base model was built using one hidden layer and 10 neurons. The paper analyzes the influence of various factors, namely the number of hidden layer neurons, the number of layers, the number of iterations, various optimizers, and activation functions on the model's performance. The experimental data for analysis in all the considered variants were divided in the proportions of 70%, 15%, 15% for training the neural network, its validation, and testing, respectively. Three variants of models using different data as input neurons are considered. As a result, the root mean square error for the test part of the data set varies from 0.05 to 0.949 °C. The mean bias error of the predicted data from the experimental data was found in the model using the air inlet temperature of the air-soil heat exchanger, humidity, and outlet temperature for the previous hour, and it was -0.001 °C. Testing of the proposed models showed that the artificial neural network predicts the temperature at the outlet of the air-soil heat exchanger, taking into account the influence of weather conditions, with satisfactory accuracy. Accordingly, artificial neural networks can be used to predict the thermal state of an air-soil heat exchanger, such as a heat pump, for which weather conditions are one of the main determining factors. In order to set up heat pump systems, experimental data are needed to describe the thermal state of shallow soils and the actual geology.