Design and optimization of the novel thermally regenerative electrochemical cycle power device based on machine learning.

• The study used CFD and machine learning to optimize TREC power device. • The trained network can forecast TREC device performance well. • The heat to electricity efficiency of TREC device can be up to 7.34%. Utilizing low grade heat and waste heat to generate electricity not only mitigates environmental impacts, meanwhile it enhances energy efficiency and lowers energy expenses. Thermal regenerative electrochemical cycles (TREC) can directly convert low grade heat to electricity, and it has a massive potential for low grade heat utilization. The type and design of a TREC system with practical and continuous power output are of paramount importance, and they can directly influence the performance and efficiency of low-grade heat recuperation systems. In this paper, a novel TREC system of a double layers rotating structure was proposed, and the novel optimal design approach employs CFD and machine learning for this system was presented. The results revealed that the height and number of TREC units had a significant impact on the device's performance. The heat to electricity efficiency of TREC device can be up to 7.34% (equivalent to 48.91% of the Carnot cycle efficiency) when the number of units was 12 and the height of unit was 5 mm. The results indicated that the MLP model was the most suitable, with RMSE values of 0.0077, 0.4415, and 0.5091 for heat to electricity efficiency, heat recuperation efficiency, and heat recuperation, respectively. This work established an effective approach for more rapid and precise estimation of the TREC device's performance metrics. The findings from prediction can serve as a practical guide for the design of TREC devices of various sizes. Furthermore, the method can envision the efficiency before the TREC device is manufactured in case of insufficient numerical calculations, reducing the computational and optimized costs. [ABSTRACT FROM AUTHOR]