Dynamic Reconfiguration of Photovoltaic Arrays for Laser Wireless Power Transmission via Deep Neural Networks

During the implementation of laser wireless power transmission (LWPT), various factors, such as clouds and atmospheric turbulence, can potentially result in nonuniform irradiation (NUI), leading to a power mismatch of the photovoltaic (PV) array. Dynamic reconfiguration technology which adopted in solar energy system is an effective method to improve the power output of PV arrays, but the current execution speed and accuracy cannot meet the fast response requirements of LWPT. In this paper, a novel adaptive threshold iteration algorithm for irradiance equalization (IE) based on total-cross-tied (TCT) configuration is proposed. Based on this algorithm, a large number of data samples characterized and labeled by irradiance and optimal layout are obtained in the simulation platform. The irradiance values are then converted into current values of special nodes for the training of deep neural networks (DNN), which simplifies both model complexity and hardware complexity. A theoretical analysis of the simulation was performed on a 3×4 scale PV array, followed by the design of the hardware circuit and evaluation of the experimental results with metrics. The results show that the output power enhancement of the array can reach up to 63.29% compared to TCT-type configuration. The conversion efficiency is improved by 17.21%, and the reconfiguration execution time is about 61 ms.