Energy-efficient optimization of radiant floor heating systems via random forest-based surrogate models and evolutionary algorithms

Radiant floor heating systems (RFHS) are among the modern heating solutions that provide high energy efficiency and thermal comfort due to their low-temperature operating principle. However, in systems using flooring materials such as wood and laminate flooring, numerous design and operational parameters can influence surface temperature and heat flux. This makes system design a complex optimization problem. In this study, an integrated optimization approach has been developed to increase the energy efficiency of RFHS with parquet flooring, minimizing surface heat flux while also meeting thermal comfort and material safety constraints. First, ANSYS-based thermal analyses were conducted for various water temperatures, ambient temperatures, pipe spacing, parquet thickness, and material thermal conductivity values, resulting in a comprehensive dataset. Using the numerical data obtained, Random Forest (RF) based predictive models capable of accurately estimating surface temperature and surface heat flux were developed. The trained RF models were integrated with a Genetic Algorithm (GA) to solve a highly constrained optimization problem. The results obtained showed that the flooring material has a decisive effect on system performance. Furthermore, considering that theoretically optimal solutions providing very low heat flux could lead to slow heating problems in practice, a scenario analysis was conducted based on the water temperature-ambient temperature difference. The scenario analysis results revealed that this temperature difference is a critical balancing parameter between the system's heating response and energy consumption. Thus, the study offers a comprehensive decision support framework that evaluates not only theoretical optimums but also practical operating conditions. The developed Random Forest-based surrogate model demonstrated high predictive accuracy with an R 2 value of 0.992 and a Mean Absolute Percentage Error (MAPE) value of 4.21%. Furthermore, the model’s robustness was validated against unseen literature data, yielding a high R 2 value of 0.985. Scenario-based optimization results indicated that, under defined constraints, heat flux values varied between approximately 1 W/m 2 and 110 W/m 2 depending on the temperature difference between the heating water and the ambient environment. Multi-run robustness analysis confirmed 100% feasibility and stable convergence of the RF-GA framework, while comparison with Random Search highlighted substantial gains in computational efficiency and constraint satisfaction. Consequently, this study combines ANSYS-based numerical analyses, machine learning-supported predictive models, and evolutionary optimization techniques to propose a rapid, reliable, and energy-focused design approach for parquet-covered radiant heating systems.