Smart grid stability prediction using artificial intelligence: A study based on the UCI smart grid stability dataset

• Published in: Sustainable computing informatics and systems Vol. 47; p. 101175
• Main Authors: Wang, Xuan, Zhang, XiaoFeng, Zhou, Feng, Xu, Xiang
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.09.2025
• Subjects: Bayesian optimization; Smart grid; Stability prediction; Support vector machine; UCI smart grid stability dataset; UCI smart grid stability dataset; Support vector machine; Stability prediction; Bayesian optimization; Smart grid
• ISSN: 2210-5379
• DOI: 10.1016/j.suscom.2025.101175

Maintaining the stability of smart grids (SGs) helps ensure that power systems continue to function well and without interruption, as renewable sources and variable demand rise. Conventional ways of monitoring tend to miss the first signs of instability, prompting the need for more intelligent solutions. This work studies the employment of machine learning (ML) to help classify and forecast SG stability, aiming to improve reliability and systems’ operational efficiency. Six algorithms, Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR), and Categorical Boosting (CatBoost), were tested using such robust metrics as accuracy, precision, recall, F1-score, ROC AUC, Log Loss, Cohen Kappa, and Matthews Correlation Coefficient. Performance of the models was increased by using GridSearchCV and Bayesian Optimization (BO) techniques. The finding is that BO-SVM achieved the highest accuracy, precision, recall, F1-score (all by 96.00 %) as well as greatest balanced accuracy and surpassed all the other methods investigated. Moreover, CatBoost and XGBoost had also steady and effective results when used with both optimization techniques. On the other hand, KNN exhibited overfitting and LR failed to capture stability patterns. These results prove that optimized SVM models are very useful for real-time monitoring of superconductor stability. Such models help make wise and prompt decisions which leads to stronger resilience in the smart grid and efficient energy use. Deploying these models under real-time, noisy, and dynamic grid environments for broader applicability would be more beneficial. •Utilization of ML-based techniques to develop predictive models to identify the stability state of smart grids.•Conducting the training and testing of the predictive model using the UCI smart grid stability dataset.•Performing the hyperparameter tuning through GridSearchCV technique.•Achieving the great efficiency of the CatBoost with the numerous accuracy and precision values of 0.944333, and 0.924905.•Providing competitive performance of XGBoost, due to their ability to understand complex relationships in the data.