SVM-Plus: An Improved Machine Learning Classifier to Predict Loss-of-Coolant Accidents in Pressurized Water Reactor

In pressurized water reactors (PWRs), it is essential to make appropriate predictions about loss-of-coolant accident (LOCA) to continue operating safely and effectively. LOCA prediction in a PWR is very complex due to the dynamics and complicated interactions present. Prediction models currently in use are not as accurate, reliable, or timely as needed for efficient decision-making and risk reduction. However, the need for improved model interpretability, the exploration of diverse machine learning (ML) models, the incorporation of real-time data for dynamic prediction, and a deeper understanding of model uncertainty in LOCA scenarios are crucial. To address these problems, this research investigates an improved ML model known as SVM-Plus, which predicts the LOCA in PWRs with more efficiency. Firstly, the dataset is collected from a marine PWR simulator containing 3668 samples. Moreover, we generated 1500 samples using generative adversarial networks (GANs). Some data preprocessing steps (cleaning, integration, clustering, etc.), are utilized, and our proposed hybrid model (SVM-Plus) is applied with 30% testing data. Lastly, the outcome of this hybrid model is evaluated using a confusion matrix. We obtained 99.613% accuracy, 0.986 recall, 0.00386 mean absolute error (MAE), and 0.00387 normalized mean absolute percentage error (NMAPE) for predicting LOCA in PWR. SVM-Plus is a new classifier for predicting LOCA in PWR. The adaptability of the SVM-Plus model is strengthened by utilizing a diverse dataset, uncovering intricate patterns and relationships beyond traditional SVM capabilities. However, real-world validation necessitates evaluating the model with actual observations from a PWR to ascertain its performance.