Adaptive rapid neural optimization: A data-driven approach to MPPT for centralized TEG systems

Abstract

Thermoelectric generation (TEG) system is designed to recover waste heat generated in industrial production and daily life with disadvantage of low power generation efficiency. Since non-uniform temperature distribution (NTD) easily gives rise to multiple local maximum power points (LMPPs) for centralized TEG systems, conventional maximum power point tracking (MPPT) methods is likely to lead to a low-quality LMPP. Hence, a novel adaptive rapid neural optimization (ARNO) approach is proposed to capture maximum power point (MPP). Particularly, generalized regression neural network (GRNN) is devoted to construct a proper mapping between control input of duty cycle and power output of TEG system, thus ARNO can implement an efficient search for MPP. In order to reduce possible instability caused by manual tuning, Bayesian optimization (BO) is adopted to adaptively select optimal parameters of GRNN. Compared to conventional MPPT methods, four simulations are carried out to verify the feasibility and advantages of ARNO, start-up experiment, step temperature variation, stochastic change of temperature, and sensitivity analysis included. Further, dSpace platform based hardware-in-the-loop experiments are performed to verify the feasibility of the proposed method.