In highway’s service areas, the charging load of electric vehicles can vary. Accurately predicting this variable load is crucial for maintaining the reliable operation of the power system. Traditional forecasting models often struggle to accurately describe the complex trends and nonlinear relationships within variable load. To address this issue, a large language model (LLM) is proposed. The LoRA parameter fine-tuning method is adopted to solve the challenge of excessive training parameters and give full play to the advantages of LLM in processing large amounts of data. In experiments, the LLM, named Llama3-8B and introduced by Meta Laboratory, has demonstrated excellent predictive ability. Its exhibits outstanding learning ability is excellent in the charging load data sets of three different service areas. The evaluation index value exceeds the comparison model by more than 60%, highlighting the potential of the Llama3-8B for variable load forecasting applications.