The present study deals with the intelligent forecasting of the coefficient performance (COP), exergy efficiency (ηex), and capital cost of a transcritical CO2 direct expansion heat pump using a heuristic machine learning approach. For the first, the parametric analysis was conducted on six important variables to examine the influence of each on ηex, COP, and the length of ground heat exchanger (LGHX). The outputs were used to develop a predictive model using the Radial Basis Functions (RBFs) artificial numeral network machine learning approach. The results demonstrated that LGHX is significantly diminished by 70.21 % as the temperature difference between the soil and evaporator increases from 5 ◦C to 13 ◦C. Also, LGHX is reduced by 21.69 % as Tin,w is intensified. Moreover, an 80 % increase in Tin,w leads to 50.12 % and 45.55 % improvement in ηex and COP, respectively. In addition, the RBF machine learning approach exhibited a high performance and accuracy in forecasting ηex, COP, LGHX, heating energy (QGC), capital cost, and the compressor and expander work with R2 ≥ 0.9943 and a low RMSE. Moreover, the results of multi-objective genetic algorithm optimization showed that the heat pump with LGC, LGHX, and compressor speed of 10.06 m, 172.5 m, and 2158 RPM brings the optimum ηex and COP of 34.22 % and 3.05, respectively, delivering heating energy of 10.3 kW with a total capital cost of 22,303 $.
