Advanced machine learning techniques: Forecasting thermal resistance in borehole heat exchanger system through RSM and hybrid DFNN-GA approaches

Abstract

The thermal efficiency of ground heat exchanger (GHEs) relies on optimizing borehole thermal resistance and total internal thermal resistance for cost-effective and efficient design. This study aims to establish a robust correlation among various parameters, including thermal conductivity of pipe, grout, and ground, borehole depth, borehole and pipe diameter, pipe-pipe distance, fluid velocity, and heat transfer rate of the ground, with the 3D borehole thermal resistance (Rb,3D) and total internal thermal resistance (Ra,3D). This exploration of eight influential factors results in an l-130 design matrix with 130 data points. A 3D computational model, utilizing a four-resistance method, accurately calculates Rb,3D and Ra,3D, while analyzing heat transfer in both solids and fluids. Two prediction models, Response Surface Methodology (RSM) and a hybrid Deep Feedforward Neural Network-Genetic Algorithm (DFNN-GA), establish the correlation between Rb,3D and Ra,3D and the eight variables. The results highlight the efficacy of the prediction models, achieving R2 values of 79.87% and 87.90% for RSM and 94.00% and 99.15% for DFNN-GA, corresponding to the predictive accuracy of Rb,3D and Ra,3D, respectively. Furthermore, lower RSME, MAE, and AARE values are observed. A rigorous comparative analysis assesses the relative effectiveness of these methodologies in achieving the research goals.