A novel intelligent framework for harnessing underground thermal energy through the optimization of grout and backfill thermophysical properties.

Borehole heat exchanger-assisted underground thermal energy storage systems are increasingly employed to enhance the reliability of renewable and geothermal energy supply, yet their operational effectiveness is constrained by complex and nonlinear interactions between subsurface formations and engineered filling materials. This study presents an intelligent, integrated framework for the prediction, optimization, and prioritization of UTES configurations. The proposed approach couples machine-learning surrogate models with evolutionary multi-objective optimization and decision-analysis techniques to address the computational limitations of conventional simulation-based design. A numerical database derived from a finite-element representation of a single U-tube borehole embedded in a geological medium is used to train the models. Five governing parameters (rock thermal conductivity, heat capacity, and density, along with grout thermal conductivity and heat capacity) are selected as inputs, while supplied and recovered thermal energies are adopted as competing performance metrics. Multilayer perceptron neural networks optimized using genetic and cheetah-based algorithms demonstrate excellent predictive capability (R ≈ 0.99). These surrogates are embedded within a multi-objective artificial vultures optimization algorithm to construct a smooth and physically consistent Pareto front spanning supplied energies of approximately 1.27-2.29 GJ and recovered energies of 0.43-0.74 GJ. The results highlight the dominant influence of thermal conductivity, particularly that of grout, on energy recovery, with diminishing performance gains observed at extreme parameter values. Finally, the COPRAS method is employed to extract seven representative design alternatives reflecting different operational priorities.