Time series analysis is considered extremely important in the industrial field, anomalies in time series can be accurately identified to effectively improve production efficiency and reduce costs, and for electricity usage, operating costs of power companies are reduced and service quality is improved through anomaly detection in users' electricity consumption time series. With the rapid development of large language models, they are being applied to more fields, including time series prediction. A method for anomaly detection of users' electricity consumption time series based on large language models is proposed, communication with large language models is achieved via prompt engineering, the time series is preprocessed so that input data can be understood by the models and the anomaly detection task can be correctly executed, and experimental analysis on real user electricity consumption data shows that abnormal data can be effectively detected by the proposed method compared with existing models.