A hybrid approach to wind power intensity classification using decision trees and large language models

Abstract

This paper proposes a Machine Learning (ML) based classification for wind power density to develop a model that can balance high accuracy with explainability for effective wind energy utilization. The proposed model achieved higher accuracy compared to traditional methods such as Weibull distribution and classical ML models, confirming the superiority of the DT-LLM hybrid approach. The dataset is the daily average of meteorological parameters, including wind speed, temperature, pressure, and air density, that will comprehensively analyze the factors of wind power density. These meteorological data were preprocessed in a structured manner to create features for use as inputs to the DT models. Their performances were evaluated based on the ROC curve, Confusion Matrix, and other metrics. LLM helped calculate and interpret Shapley values, which enhanced the model's explainability. The main findings include identifying wind speed at 50 m above ground (DAWS50) as crucial for model performance. This study will provide a high-performance, interpretable framework that will help overcome the limitations of traditional models in not only classifying wind power density and explaining it to enhance its applicability in decision-making. The obtained results will improve the process of modeling renewable energy more effectively and further guide researchers toward a better vision in this direction. To the best of our knowledge, this is the first study to combine Decision Trees and Large Language Models for WPD classification, providing a novel balance between model accuracy and interpretability.