基 于 大 语 言 模 型 的 电 力 系 统 通 用 人 工 智 能 展 望: 理论与应用

The large language model (LLM) is a deep learning language model that utilizes large-scale text corpora for pre-training and fine-tuning. Nowadays, it has demonstrated powerful capabilities in generalized quizzing, text generation and scientific reasoning. In this context, this paper explores the construction of artificial general intelligence techniques for power systems based on LLM and prospects its potential applications in power systems. Firstly, the basic principles, neural network architecture, and training methods of LLM are introduced, with a particular focus on its breakthroughs in logical reasoning, programming and code understanding, and mathematical reasoning compared with traditional artificial intelligence models. Then, this paper prospects the potential applications of LLM in the areas of load forecasting and renewable energy generation prediction in power systems, power system planning, power system operation, fault diagnosis and system restoration in power systems, and electricity markets. Finally, the challenges in building an artificial general intelligence technology for the power system based on LLM are elaborated upon, including data quality and accessibility in the power system domain, interpretability of output results, and privacy protection concerns.