Power grid-connected photovoltaic system voltage stability evaluation using machine learning

The integration of PV systems into electrical grids affects the voltage stability of these systems due to the intermittency of solar production and the inability of photovoltaic energy systems to supply reactive power. Indeed, the variation in reactive power margins of different load buses is one of the consequences of this integration. However, to evaluate this margin and identify sensitive areas, the appropriate method is the Q-V curves. Nonetheless, this method incurs costs in terms of computational complexity and robustness. For this reason, machine learning models are utilised. The objective of this work is to predict the voltage stability margin of a network based on Q-V analysis through the integration of photovoltaic systems into the grid. Thus, four machine learning models are proposed: support vector machines, decision trees, random forests, and artificial neural networks. Tests conducted on the IEEE 14-bus system demonstrated, by integrating the photovoltaic source into various load nodes with penetration rates ranging from 0% to 100%, that artificial neural networks were more accurate, considering their performance across the training, testing, and validation datasets.