K-Nearest Neighbors hybrid method for maximum power point tracking under partial shading for photovoltaic power systems
Djamel Guessoum, Maen Takruri, Mohammad Rabih, Maissa Farhat, Sufian A. Badawi
Abstract
Partial shading conditions (PSC) can significantly reduce the performance of solar photovoltaic (PV) systems. Consequently, to enhance maximum power point tracking (MPPT) under such conditions, the implementation of advanced algorithms is imperative for effectively addressing the challenges posed by multiple local power peaks. This paper presents a hybrid approach that integrates the K-Nearest Neighbors (KNN) machine learning algorithm with an enhanced local search for optimizing the duty cycle D. This method is inspired by the Perturb & Observe (P&O) algorithm. The KNN algorithm serves as an initial step to identify the irradiance levels around the actual solar irradiation, along with their corresponding maximum power ( ) and maximum voltage ( ). Using these values in conjunction with the load, a set of duty cycles ( ) is calculated, creating a limited search space for the optimal duty cycles. The collected dataset inputs are the non-uniform solar irradiance levels ranging from 100 W/m² to 1000 W/m², incrementing by 50 W/m². It includes all possible combinations of irradiation conditions, with the outputs being the and the corresponding . The proposed method has surpassed various optimization techniques for MPPT, including particle swarm optimization (PSO) and Cuckoo search (CS). It achieved an impressive efficiency of 98.227%, enhanced stability around the maximum power point (MPP), and a fast tracking speed of 3.9635 ms. Furthermore, this method has been used to initialize PSO, CS, and P&O algorithms, significantly improving their overall performance under PSC. Moreover, the KNN method proved to be highly adaptable to variations in PSC and rapid changes in solar irradiation.