RBMET: A New Machine Learning Approach for Predicting Grid Stability in the Electric Grids

Grid stability is crucial for meeting the demands of a modern, evolving power system, providing a dependable power supply, enhancing system resilience, enabling effective energy management, encouraging the integration of renewable energy, and reducing cascade failures to preserve the faultless operation of cutting-edge machinery, monitoring systems, and control mechanisms. In this research, the prediction of electric grid stability is performed with a machine learning (ML) approach. The dataset is obtained from Kaggle and comprises 60000 data and 14 features. Again, 20000 more data are generated using Automated Machine Learning (AutoML). Data pre-processing is done, which enables ML models to make predictions that are more accurate. As the dataset contains a group of unlabeled data, k-means clustering is executed. Dynamic feature selection is performed with Restricted Boltzmann Machine (RBM) and stability in electric grids is detected with Extra Trees. Extra Trees combines numerous decision trees to generate predictions and classifies them into different categories of instabilities, such as voltage instability, frequency instability, and power imbalance. This combination of RBM and Extra Trees is defined as RBMET in this research. A confusion matrix is generated to determine accuracy, precision, and recall. Training time is also generated for performance analysis. This RBMET model demonstrates 97.39% accuracy score, along with precision (0.974), and recall (0.974). Moreover, 23.19 seconds of training time are required to train the entire model. This research can be helpful for engineers and power grid operators to take immediate steps to prevent any kind of failure in the electric grids.