Performance prediction and optimization of annular thermoelectric generators based on a comprehensive surrogate model

Abstract

A traditional system-level thermoelectric model requires enormous computing power and time for simulation analysis, especially when multiple optimization algorithms are combined. This paper initially proposed a comprehensive surrogate mode for the accurate modeling, field analysis, and fast optimization of thermoelectric generators. The surrogate model is made up of an artificial neural network (ANN) and a conditional generative adversarial network (cGAN), which can achieve a prediction accuracy of 97 % and a structural similarity (SSIM) of 0.954. Using a twisted tape annular thermoelectric generator (TT-ATEG) as an example, the generalization capability of the comprehensive surrogate model is demonstrated by studying the change of the twisted tape geometry parameters on the output performance and field distribution of the annular thermoelectric generator. The combination of the surrogate model and NSGA-II achieves fast optimization of key parameters under different working conditions, and the computational efficiency is improved by 99.97 %. Meanwhile, the cGAN part can predict the pressure and heat flow fields within 5s to provide intuitive visual feedback. The trained surrogate model requires less computer arithmetic power compared to the traditional multi-physics field simulation software. The successful application of the comprehensive surrogate model in this work provides a new solution idea and an optimization method for system-level TEG.