Enhancing detection of THD voltage events in solar power quality data using meta-learning and standard thresholds

The integration of solar farms into existing power grids presents numerous challenges, particularly in terms of power quality and grid stability. This paper investigates the detection and classification of risky events in Total Harmonic Distortion of Voltage (THD-V) within a solar farm located in Norfolk, England. Using data collected from January to December 2023, we apply the IEEE 519-2014 standard to identify risky events where THD-V exceeds 5%, and classify them as critical disturbances. A threshold-based classification approach is combined with meta-learning-based machine learning, specifically Model-Agnostic Meta-Learning (MAML), to detect these events. The model is trained on data of 2023, employing the Random Oversampler technique to address the imbalance in risky event occurrences. This research provides insights into how these events can impact grid reliability and lead to equipment failures or operational inefficiencies in solar farms. Our results demonstrate the effectiveness of machine learning in improving early detection and classification of risky events, thereby enhancing the resilience of power systems integrating renewable energy sources.