Machine Learning-Enabled Optimization Framework for Application-Oriented Aqueous Thermocell Units and Array Modules for Low-Grade Heat Harvesting

The efficient conversion and utilization of low-grade thermal energy is essential to enhancing the energy efficiency and reaching the carbon neutrality goals of modern industrial energy systems. As a cutting-edge heat-to-electricity technology, thermo-electrochemical cells (also known as thermocells) have great prospects in low-grade heat recovery due to its ultra-high thermopower, high scalability, and low cost. However, conventional aqueous thermocells, typically based on planar or rectangular modules, face substantial limitations when deployed in real-world environments characterized by geometric irregularity, spatial constraints, and distributed thermal sources. This study addresses the utilization of waste heat in complex circular tubular sources by proposing a machine learning-enabled optimization framework for fan-shaped aqueous thermocells. Firstly, a database is established using simulation models. By integrating machine learning methods, a surrogate model is developed to achieve high-accuracy performance prediction of fan-shaped thermocells under various geometric configurations and deployment positions. Secondly, by using the surrogate model to decouple the relationship between the geometric parameters and the deployment position, a two-step Gray Wolf optimization algorithm (GWO) is proposed to optimize the performance of single fan-shaped thermocell unit. Its maximum power density can reach 333.7 mW·m−2, which is 66.35% higher than that of the conventional square one. To further conquer the issue of full coverage deployment, a synergistic performance optimization for array-level fan-shaped thermocells modules (hereinafter referred to as the “module”) is proposed. Through a combinatorial arrangement, six fan-shaped thermocell units are aligned to conform to circular heat sources, resulting in a power density of 235.7 mW·m−2. These results highlight the potential of geometry-driven and machine learning-assisted design in advancing next-generation thermocells technologies toward scalable applications.