This research presents an innovative approach to bolster Smart Grid stability prediction through the optimization of the Extreme Learning Machine (ELM) model using both Bayesian Optimization (BO) and Particle Swarm Optimization (PSO). The study compares various predictive models, including Logistic Regression, Decision Trees, XGBoost, Random Forest, ELM, BO + ELM, and the proposed PSO + ELM. Notably, the PSO + ELM framework emerges as the most effective, achieving unparalleled precision and accuracy at 98.45%, surpassing traditional and contemporary machine learning methods. The study utilizes real-world smart grid stability data from a four-node architecture, providing a robust evaluation of operational complexities. Results affirm the reliability and effectiveness of the PSO-optimized ELM, showcasing superior predictive performance across diverse metrics. The research contributes to optimizing modern power systems, enhancing reliability, and resilience. Future investigations may explore the scalability and adaptability of the proposed PSO + ELM framework to larger and more complex Smart Grid architectures, as well as its applicability to real-time data streams and dynamic system conditions for further practical utility.