Adversarial Machine Learning Architecture in AI-Driven Power Systems

The paper investigates the use of AI in power systems to improve monitoring, control, and optimization. For successful classification tasks on various datasets, ML models are critical, boosting power system dependability and efficiency. The research focuses on ML model designs in AI-driven power systems for binary, triple, and multiclass classification. Misclassifications can destabilize grid stability, underlining the importance of precise classification in decision-making processes. To combat this, this research proposes a DNN model that uses specific parameters that were fine-tuned for greater accuracy in classification tasks. Model correctness is an important performance metric that can be improved by architecture design and fine-tuning. For successful optimization, three essential loss functions (binary cross-entropy, categorical cross-entropy, and sparse categorical cross-entropy) are investigated. We have achieved great results with utilizing the sparse categorical cross-entropy loss function to minimize the error during training periods. The study underscores the critical relevance of correct categorization and the possibility for robust AI applications in power systems, which contribute to grid stability and sustainability.