Predicting Renewable Energy Using Multi-Agent Large Language Models

With the increasing adoption of renewable energy, accurate forecasting of wind and photovoltaic (PV) power generation is essential for optimizing power system dispatch and ensuring grid stability. This paper proposes a joint forecasting method for wind and PV power generation based on large language models (LLMs). The method incorporates both a Single-ChatGPT-Agent model and a Multi-ChatGPT-Agent model, each designed to address the limitations of traditional forecasting approaches in handling renewable energy uncertainties by enhancing prediction accuracy. Comparison with a classical benchmark method demonstrates that the Single-ChatGPT-Agent model significantly outperforms the classical benchmark method in terms of forecasting accuracy for both wind and PV power generation. Furthermore, the Multi-ChatGPT-Agent model exhibits significantly lower error compared to the Single-ChatGPT-Agent model in both wind and PV forecasting experiments. Additionally, the overall prediction error of the Multi-ChatGPT-Agent model is substantially reduced compared to both the classical benchmark method and the neural networkbased predictions, highlighting the advantages of this approach in improving prediction accuracy and effectively addressing renewable energy uncertainty. The results indicate that the LLMbased prediction method, especially the Multi-ChatGPT-Agent approach, offers superior ease of operation, greater prediction stability, and enhanced forecasting accuracy, making it an innovative solution for optimizing renewable energy integration and improving system reliability.