A Causality-Oriented Neural Network Framework for Transient Stability Prediction in Power Systems

Transient stability prediction is crucial for dynamic security assessment in power systems. Conventional correlation-based Machine Learning (ML) approaches, while computationally efficient, often lack interpretability and fail to capture the true causal relationships among system variables. This paper presents a causality-oriented Artificial Neural Network (ANN) framework designed to address these limitations. First, a hybrid causal discovery method is proposed to identify physically meaningful causal relationships among system variables, enabling a principled selection of features that directly impact transient stability. Second, these causal insights are systematically integrated into ANN training through optimized feature selection and weighting schemes. The case study results verify that the proposed framework achieves superior generalization, reduces overfitting, and improves both predictive accuracy and interpretability across key evaluation metrics. By integrating causal discovery with neural network learning, the proposed approach offers a more robust solution for transient stability assessment and advances causality-aware ML applications in power system security.