Novel predictive modeling of capacitor module activation in critical power systems using random forest and SMOTE: A case study at a radioactive waste management installation

Capacitor banks play a vital role in modern power systems by improving power factor and minimizing energy losses. Their reliable operation is essential for system stability and energy efficiency, especially in industrial and high-demand environments. While machine learning is increasingly used in power system optimization, predictive modeling of capacitor module status remains a relatively underexplored area, particularly in complex, mission-critical settings. This study proposes a novel application of the Random Forest algorithm, enhanced with the Synthetic Minority Oversampling Technique (SMOTE), to predict the activation status of 12 capacitor modules in a 500 kVAR bank at the Radioactive Waste Management Installation (RWMI). In contrast to traditional approaches focused on placement or capacity optimization, our method utilizes real operational data and explicitly addresses severe class imbalance, a common yet often neglected issue in predictive modeling. By integrating SMOTE, the model achieves a balanced classification performance with a prediction accuracy of up to 96%, along with high precision, recall, and F1-score. These results confirm the effectiveness of the proposed method in supporting more informed operational decisions and improving power system stability. The study highlights the value of data-driven approaches for optimizing energy management in critical infrastructure.