As the agricultural load is greatly affected by meteorological factors and a single decomposition method cannot fully extract the multidimensional features existing between multiple inputs, an agricultural load forecasting model based on multivariate variational mode decomposition combined with SVR Bi GRU TCN combined model is proposed. Firstly, using multivariate variational mode decomposition to adaptively decompose historical agricultural loads and meteorological characteristics, real-time mining of modal components with different feature scales between data is carried out. Then, based on the inherent properties of each modal component, SVR, Bi-GRU, and TCN models are established to extract feature information at different time scales, thereby achieving accurate prediction of future 1-hour agricultural loads. The experimental results show that compared with the SVR model, Bi-GRU model, and TCN model, LSTM model and CNN-BiLSTM model, the proposed prediction model can effectively improve the prediction accuracy.