Identification of Solar PV Array Partial Shading Patterns using Machine Learning
Mismatch losses due to partial shading can limit the energy generation of solar photovoltaic (PV) systems. Isolating the shaded PV modules through electrical reconfiguration can potentially improve the power output of the PV array. To do this, the shaded modules need to be identified before the PV array can be reconfigured to produce the optimum output power. In this study, an algorithm was developed to identify the partial shading pattern of a PV array using machine learning. Measurements from the current sensor integrated into the switching circuit of each module and the solar irradiance from a pyranometer were utilized as input to the machine learning algorithm. The algorithm was trained using the voltage and current readings of an off grid PV system composed of nine 10-W PV modules arranged in a 3 × 3 array in series-parallel configuration. Three machine learning techniques were used, namely SVC, Random Forest, and K-Nearest Neighbors, resulting in 80 %, 86 %, and 66 %, respectively, in terms of accuracy, precision, recall, and f-1 score. Thus, the Random Forest algorithm was found suitable for this type of problem as it can reliably distinguish the shading patterns on the array.
N. G. T. Ramirez and E. Q. B. Macabebe, "Identification of Solar PV Array Partial Shading Patterns using Machine Learning," 2019 IEEE PES Asia-Pacific Power and Energy Engineering Conference (APPEEC), Macao, Macao, 2019, pp. 1-5. doi: 10.1109/APPEEC45492.2019.8994360