Explainable and counterfactual lasso regression for resilient micro gas turbine power prediction in smart grids

Abstract

Accurate prediction of electrical power output from micro gas turbines is essential for optimizing performance in microgrids and distributed power systems. This study introduces a novel and interpretable machine learning framework using Lasso regression applied to a newly published dataset titled Micro Gas Turbine Electrical Energy Prediction, available on Kaggle. The dataset captures time-series relationships between input control voltage and electrical power output, enabling effective modeling of micro turbine behavior. The proposed model relies on only two features, input voltage and time, ensuring computational efficiency while maintaining predictive performance. To support decision-making and model transparency, the framework incorporates Explainable AI (XAI) techniques such as SHAP and LIME, which reveal the influence of input features on predictions. Additionally, counterfactual analysis is integrated to explore how changes in inputs affect predicted outcomes. This allows users to define a minimum and maximum range for desired power outputs, providing actionable insight. The approach demonstrates high accuracy, with over 87 % of predictions falling into the low or category. By enabling interpretable and resource-efficient forecasting of local energy generation, the proposed framework contributes to the development of resilient and sustainable smart grid infrastructures. Most importantly, the proposed system is highly relevant for smart grid and microgrid operations, where transparent, accurate, and adaptive prediction of local generation units like micro gas turbines plays a critical role in maintaining system stability, load balancing, and energy efficiency.

Keywords: Explainable AI (XAI), Lasso Regression, Machine Learning Framework, Resilient Cyber-Physical Systems, SHAP and LIME, Sustainable Smart Grids.