Enhancing Solar Power Generation Through Threshold-Based Anomaly Detection in Errachidia, Morocco

This research presents an innovative approach to optimize solar power generation in arid regions, focusing on Errachidia, Morocco. The study effectively detects anomalies in solar radiation data by leveraging machine learning techniques such as Isolation Forest (IF), Local Outlier Factor (LOF), and Principal Component Analysis (PCA). These anomalies, driven by dust storms and equipment malfunctions, are classified by severity through threshold-based methods (TBM). The results are promising, with Area Under the Precision-Recall Curve (AUC-PR) scores of 0.8423 for IF, LOF, and PCA. While DBSCAN yields a lower score of 0.0006, the model's overall accuracy is impressive at 99.74%. Robust cross-validation results demonstrate the model's reliability, with a mean accuracy of 99.81%. Notably, feature selection using Random Forest (RF) highlights critical variables, including High Speed (Hi Speed), Outdoor Humidity (Out Hum), Barometric Pressure (Bar), Indoor Humidity (In Hum), Indoor Electromagnetic Compatibility (In EMC), High Solar Radiation (Hi Solar Rad.), Dew Point (Dew Pt), Low Empowerment (Low emp), and Indoor Dew Point (In Dew). The study underscores the potential of machine learning and TBM approaches in solar energy management and offers a replicable framework for regions facing similar climatic challenges.