OptRAAL: A Robust Machine Learning Classifier for Predicting Anomalies in Fusion Power Reactors

Fusion reactors offer clean energy but face challenges in anomaly detection for safety and performance. Hence, fusion power research needs advanced machine learning (ML) methods to enhance anomaly detection and predictive maintenance in nuclear fusion reactors. In this study, a novel ML framework, Optuna Randomized Adversarial Adaptive Learning (OptRAAL) is introduced, integrating Random Forest, AdaBoost, Bagging, and a meta-learner optimized through Optuna. By leveraging stacking and voting mechanisms, OptRAAL ensures improved accuracy, robustness, and adaptability in detecting critical anomalies such as plasma disruptions, superconducting quenching, and fuel contamination. The balanced dataset, containing 100,000 fusion experimental data, includes nineteen critical reactor parameters, with major magnetic confinement configurations (Tokamak, Reversed Field Pinch, and Stellarator). To enhance clustering accuracy, k -Means ++ was employed for better centroid initialization. The OptRAAL framework combines ensemble techniques with fusion-specific feature engineering, an adaptive loss function, and dimensionality reduction to handle high-dimensional reactor sensor data efficiently. This ensures precise classification while mitigating noise and uncertainty. Comparative benchmarking against several regular ML models demonstrates its superior $99.667 \%$ accuracy, 0.9967 precision, and 0.9967 recall, with an average inference time of 0.041 milliseconds per sample. By significantly improving fault detection, predictive maintenance, and real-time adaptability, OptRAAL enhances fusion reactor safety and efficiency. Its scalability and robustness make it a promising solution for continuous monitoring, contributing to the long-term viability of fusion energy as a clean, carbon-free power source. The robustness, interpretability, real-time processing, efficacy, feature engineering, and deep learning integration can be enhanced for improved anomaly prediction in fusion reactors.