A hybrid machine learning, metaheuristic optimization, and decision-making framework for solar chimney ground material selection considering radiation effects

A unified, data-driven framework has been developed to address a key methodological gap in the optimization of solar chimney power plants (SCPPs). This framework integrates surrogate modeling, multi-objective optimization, and multi-criteria decision-making to maximize updraft velocity (UV), temperature rise (ΔT), and collector efficiency (CE) by systematically tuning ground thermophysical properties under realistic radiation and environmental conditions. A four-stage pipeline was implemented: data preprocessing and exploration, GMDH-type neural network modeling, Pareto-based optimization using NSGA-II, and final scenario ranking through TOPSIS. High predictive accuracy was achieved by the ANN-based surrogate models (R > 0.9994). Through optimization, key performance regions were identified: updraft velocities up to 14.6 m/s were obtained under low thermal conductivities and high solar radiation, collector efficiency reached 39.8 % at moderate radiation levels, and temperature rises of 28.4 K were achieved in cold ambient conditions with strong radiation exposure. Semi-conductive soils (0.3–0.6 W/m·K) and medium densities (1650–1800 kg/m 3 ) were selected in 55 % of Pareto solutions, while 77 % favored high-radiation settings. Practical scenarios were derived, enabling rapid customization of SCPP designs without re-optimization. This integrated approach is expected to reduce trial-and-error, clarify design trade-offs, and accelerate context-sensitive deployment and policy integration of radiation-based renewable energy systems.