Machine learning-based voltage fault ride-through assessment of inverter-based resources using phasor measurement units

This paper presents a methodology for assessing the fault response performance of inverter-based resources (IBR) in the National Interconnected System of Colombia, using artificial intelligence algorithms and phasor measurement data. The methodology analyzes the voltage behavior during faults, verifying whether the IBRs comply with the grid code requirements, particularly in terms of their ability to remain connected during faults (Fault-Ride-Through, FRT). The methodology consists of two main stages: the first stage involves training artificial intelligence algorithms to classify the fault response performance of the IBRs, and the second stage focuses on performance assessment using measurements and the trained intelligent machines that classify the IBR behavior according to the expected response thresholds established in the grid code. The methodology was validated with real data from photovoltaic solar plants in Colombia, using information from the FRT curves and the requirements of the Colombian grid code. Algorithms such as Random Forest, Support Vector Machine, and Neural Networks were tested. The results show that the methodology can assess the performance of the IBRs and detect noncompliance with the FRT curves. The random forest regression algorithm performed best, with a training regression error of less than 0.22% and correctly classifying 100% of the analyzed cases. This methodology allows for the definition of improvement actions in the protection and control systems of generation plants, ensuring greater resilience and stability in the electrical grid.