Accurate short-term load forecasting of rural integrated energy systems is of great significance for the safe and stable operation of rural power grids and local renewable energy consumption. However, due to the coupled effects of agricultural production rhythms, sudden weather changes, and other factors, the load of rural integrated energy systems exhibits strong randomness, non-stationarity, and multi-time-scale characteristics, which limit the forecasting accuracy of traditional methods. To address this issue, a short-term load forecasting method is proposed based on the VMD-Informer-BiGRU model. Firstly, a multidimensional feature set including farming periods, ambient temperature, and holidays is constructed to accurately characterize the unique energy consumption patterns in rural areas. Variational mode decomposition (VMD) is then used to decompose the original load sequence into modal components with different frequency characteristics, effectively reducing random fluctuation interference. On this basis, the Informer model is introduced to extract global long-term temporal dependencies, while the bidirectional gated recurrent unit (BiGRU) is employed to capture local mutation features, achieving deep fusion of multi-scale temporal features. Finally, a combined forecasting model is established. Using an actual rural energy consumption scenario in China as the case study, the results show that the proposed model achieves higher forecasting accuracy than other models.