Review of Power System Prediction Technologies Based on Large Language Models

ObjectivesWith the large-scale integration of renewable energy, new-type power systems require greater flexibility and higher prediction accuracy for prediction technology. Traditional prediction methods have limitations in handling dynamic and complex scenarios, highlighting the need for prediction technologies tailored to these systems. Large language models (LLMs), as generative artificial intelligence technologies, have capabilities in multimodal data integrating, few-shot learning, and multitask handling, enabling more intelligent and precise solutions for the prediction of power systems. Therefore, this study focuses on analyzing the current applications and advantages of LLMs in power system prediction.MethodsFirst, the fundamental architecture, training methods, and current application status of LLMs are discussed. Then, their principles and implementations in prediction are explained, with emphasis on advantages and prospects in load prediction, renewable generation prediction, and electricity price prediction. Finally, challenges in LLM-based prediction applications are analyzed from three aspects: data quality management, privacy protection, and computational resources, and feasible solutions are proposed.ConclusionsThrough comparative analysis of various forecasting tasks, LLMs demonstrate superior capabilities in few-shot learning and multimodal data processing compared to traditional methods, making them more adaptable to complex and variable prediction scenarios. Effective application of LLMs can provide innovative solutions for electricity market prediction.