关键词: 大语言模型(LLM), 人工智能技术, 负荷预测, 电价预测, 新能源出力预测, 隐私保护

Abstract: [Objectives] With the integration of large-scale new energy sources into the grid, the new power system not only demands higher accuracy in forecasting technology but also requires certain flexibility. Traditional forecasting methods have certain limitations when dealing with dynamically complex scenarios. Therefore, it is necessary to study forecasting technologies that are suitable for the new power system. [Methods] Large language model (LLM) are a type of generative artificial intelligence technology, capable of processing multimodal data, few-shot learning, and multitasking. Therefore, this paper first explains the basic architecture, training methods, and current applications of LLMs, and then describes the principles and implementation process of their application in the forecasting field. It focuses on discussing the advantages and potential of LLMs in aspects such as electric load forecasting, new energy output forecasting, and electricity price forecasting. Finally, it analyzes the existing problems of LLMs in forecasting applications from three aspects: data quality management, privacy protection, and computational resources, and provides feasible solutions. [Results] After reviewing different forecasting tasks and comparative research analysis, compared with traditional forecasting methods, LLMs' strong capabilities in data processing and multitasking are more suitable for handling complex and variable forecasting scenarios. [Conclusions] The rational and effective application of LLMs can provide new solutions for power market forecasting.

Key words: large language model (LLM), artificial intelligence technology, load forecasting;?electricity price forecasting, new energy output forecasting, privacy protection