Logistic Regression Based Decision Support System for Power Grid Operations

Abstract

Ensuring power grid stability during critical operating situations requires prompt and precise choices by system dispatchers. This paper uses Logistic Regression (LR) to forecast dispatcher actions using the Power Grid Dispatcher Operations dataset from Kaggle, including 1,000 labelled scenarios that include voltage deviations, load variations, and equipment failures. Following preprocessing and feature selection, a binary LR model was developed to classify whether dispatchers will implement remedial measures. The model achieved an accuracy of 98.50%, precision of 98.02%, recall of 99%, and an F1-score of 98.50%, demonstrating robust predictive efficacy. Critical factors affecting decisions were frequency variation, reserve margins, and line loading percentages. Class-weighted loss functions and stratified sampling were used to mitigate data imbalance among action classes. The model's interpretability, together with its efficacy, endorses its use as a decision-support instrument in real-time operations. The results indicate the effectiveness of statistical learning methods in improving dispatcher situational awareness, minimizing human error, and strengthening grid resilience under high-risk situations.