Adaptive PV Systems: Using Genetic Algorithm and Random Forest Classifier to Mitigate the Effects of Partial Shading on Solar Photovoltaic Systems

Mismatch losses due to partial shading of solar photovoltaic (PV) arrays can limit their generated power. One method to improve their power output is to electrically reconfigure the array based on the principle of row irradiance equalization. Shaded panels in an array are identified to determine the array's optimal configuration and maximize its power output. In this study, a random forest classification model was developed to identify the partial shading pattern incident on an array. Features of the model included the individual panel currents and the array irradiance of preset 20 shading patterns. The data used to train the model was obtained using an off-grid <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3\times 3$</tex> array composed of nine 10-W PV modules in total-cross-tied (TCT) topology. The model achieved a mean accuracy of 90% using the testing data and 98% using 5-fold cross-validation. This was integrated with a genetic algorithm-based irradiance equalization heuristic, which takes a predicted shading pattern as input and outputs an optimized configuration. For each shading pattern, the suggested reconfiguration was recorded and its corresponding power change was calculated. Power recovery ranged from −15.70% to 40%. The <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3\times 3$</tex> PV array reconfigured successfully to the optimum configuration.