Abstract

Anovel global maximum power point tracking technique for a partially shaded photovoltaic (PV) system based on an artificial intelligence strategy that hybridizes decision trees (DTs) with the classic perturb and observe (P&O) algorithm is presented in this article. This hybridization ensures the best tradeoff between both techniques, which means a fast response during steady-state with a low computational cost provided by the P&O algorithm and an outstanding response to variations in irradiance, even in situations of partial shading through the DT method. In this sense, the DT method requires a high demand for data with different irradiance cases for offline training. However, its implementation generates a straightforward structure that is easy to evaluate in real-time with a low computational cost, enabling deployment on low-cost microcontrollers. The proposed technique has been extensively validated with different experimental results using a PV emulator and a dc–dc power converter controlled by a low-cost microcontroller. The results have demonstrated the superiority of the proposed method over two classical techniques designed to operate under solar partial shading conditions. © 2020 IEEE.