Machine learning models for smart grid stability prediction: a comparative analysis.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• The integration of renewable energy sources and fluctuating demand challenges smart grid (SG) stability.
• Traditional methods for monitoring grid instability are often reactive and insufficient for modern energy systems.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) methodology for accurate classification and prediction of smart grid stability.
• To enhance the effectiveness and reliability of smart grid operations through intelligent monitoring.

Main Methods:

• Utilized fourteen distinct machine learning models for classification and prediction tasks.
• Applied advanced feature engineering and selection techniques (filter, wrapper, embedded) to optimize model performance and reduce dimensionality.
• Employed Bayesian optimization (TPE) and metaheuristic optimization (GWO) for hyperparameter tuning to maximize ML model accuracy.
• Incorporated explainable AI methods to ensure model trustworthiness and transparency.

Main Results:

• The Light Gradient Boosting Machine model demonstrated exceptional predictive performance, achieving near-perfect accuracy.
• The TPE-optimized model reached an accuracy of 99.95%, while the GWO-optimized model achieved 99.90%.
• Feature selection and hyperparameter optimization significantly improved model accuracy and efficiency.

Conclusions:

• The proposed ML methodology offers a robust solution for predicting smart grid stability, surpassing traditional approaches.
• The study highlights the potential of advanced ML techniques, particularly Light Gradient Boosting Machine, in improving smart grid resilience and supporting energy efficiency.
• Explainable AI integration ensures the reliability and trustworthiness of the predictive models for practical application.