This survey provides a comprehensive overview of the use of large language models (LLMs) in time series analysis. It highlights the challenges and motivations for applying LLMs to time series data, which includes the large volume and variety of data, as well as the non-stationary nature of real-world time series, leading to concept drift. The survey categorizes existing methods into five groups: direct query, tokenization, prompt design, fine-tuning, and model integration. It discusses the general pipeline for LLM-based time series analysis and provides a detailed taxonomy of these methods. The survey also reviews applications of LLMs in both general and spatial-temporal time series data, covering domains such as finance, healthcare, traffic, and computer vision. Finally, it outlines future research opportunities to further advance time series analysis with LLMs, emphasizing the need for novel tokenization methods and better prompt design.This survey provides a comprehensive overview of the use of large language models (LLMs) in time series analysis. It highlights the challenges and motivations for applying LLMs to time series data, which includes the large volume and variety of data, as well as the non-stationary nature of real-world time series, leading to concept drift. The survey categorizes existing methods into five groups: direct query, tokenization, prompt design, fine-tuning, and model integration. It discusses the general pipeline for LLM-based time series analysis and provides a detailed taxonomy of these methods. The survey also reviews applications of LLMs in both general and spatial-temporal time series data, covering domains such as finance, healthcare, traffic, and computer vision. Finally, it outlines future research opportunities to further advance time series analysis with LLMs, emphasizing the need for novel tokenization methods and better prompt design.