Predicción del Estado de Estabilidad de Corto Plazo en Sistemas de Potencia con Integración de Generación Renovable Utilizando Aprendizaje Profundo

This paper presents a deep learning-based methodology for short-term stability prediction in electrical systems with high renewable generation penetration. Based on dynamic simulations performed on the 39-bus IEEE system, including scenarios with wind and photovoltaic integration, a database was constructed to train a hybrid recurrent convolutional neural network (RCNN-EE) model. The model uses strategically selected electrical variables to capture both transient dynamics and voltage collapse phenomena, achieving accurate classification of the system state. The results obtained demonstrate superior performance compared to traditional architectures, achieving outstanding metrics even under unbalanced conditions. In addition, its applicability in real time is validated, with inference times of less than 50 milliseconds, which demonstrates its potential for implementation in electrical system protection and control schemes.