AutoFEnergy -- Synthetic Data Used to Teach LLM Feature Engineering for Energy

The synthetic AutoFEnergy dataset is the first publicly available dataset designed to train LLMs with engineer-level feature-engineering tool-calling capabilities in the energy domain, capabilities that are not inherently present in general-purpose LLMs.This work is submitted as a dataset entry to the ``Good Datasets for AI Model Training in the Power and Energy Domain'' competition, with the goal of facilitating research understanding and reproducibility.The AutoFEnergy dataset aims to advance feature-engineering tasks across a wide range of forecasting and classification scenarios in the energy domain toward LLM-based end-to-end automation, thereby reducing reliance on experienced engineers and significantly lowering both human labor costs and time expenditure.All data generation, fine-tuning, and validation procedures are implemented through reproducible, open-source Python workflows, and are accompanied by accuracy evaluations to verify the effectiveness of the dataset.