Improving predictive reliability and automation of smart grids using the StarNet ensemble model

Ensuring predictive reliability and automation in smart grids is vital for stable and efficient electricity distribution. This study introduces the StarNet Ensemble Model, a web-based intelligent framework designed to enhance grid stability through a stacking-based ensemble learning approach. The system integrates a machine learning-driven graphical user interface (GUI) that enables automated, real-time monitoring and prediction of grid performance. A synthetic dataset, generated from a four-node star network and extended with consumer node variations, was initially used for model development and achieved a preliminary predictive accuracy of 99.43%, surpassing conventional methods. To extend applicability beyond simulated data, the model was validated using two benchmark datasets, namely the UCI Smart Grid Stability Dataset and the IEEE 14-Bus Test System. The StarNet model employed CatBoost, AdaBoost, Random Forest, SVM, and KNN as base learners with a Random Forest meta-model, evaluated through stratified 10-fold cross-validation. The model achieved 98.94% accuracy on the UCI dataset and 97.83% on the IEEE 14-Bus dataset, with a cross-dataset transfer accuracy of 95.41%. These results confirm the model's robustness and generalization capability, demonstrating that the StarNet Ensemble Model effectively enhances predictive reliability and automation in smart grids.