To solve the multi-energy coupling relationship and improve the load forecasting accuracy of integrated energy system under the scarce historical data scenario

the TimeGPT integrated energy system multivariate load forecasting method based on the time-series large model is proposed

which firstly analyzes the influence of meteorological factors on the multivariate loads and introduces the distance correlation coefficient to screen the meteorological factors

and utilizes the mechanism of self-attention to capture the coupling relationship among different loads; and secondly

utilizes the pre-trained large model to feature-fuse the meteorological factors as TimeGPT exogenous variable inputs

and locally tune the parameters through the large model fine-tuning technique. Finally

the experiments show that the pre-trained and fine-tuned TimeGPT model has higher forecasting accuracy in multivariate load forecasting compared with the traditional machine learning model under scarce historical data.