Abstract: Based on the meteorological and temporal characteristics of short-term load in greenhouses, an integrated short-term load prediction architecture based on the fusion of variational mode decomposition (VMD), convolutional neural network (CNN), and long short-term memory (LSTM) is proposed to solve the problem that the electricity load of agricultural greenhouses is affected by rural power supply capacity and meteorological factors, which have strong fluctuation and high nonlinear characteristics. First, the load sequence is decomposed using the VMD method to reduce load volatility. Second, the meteorological characteristics of the load are extracted using the CNN method, and the temporal characteristics of the load are extracted using the LSTM method. Component prediction is performed, and the prediction results of the model components are reconstructed. Finally, based on the load data of agricultural greenhouses in Shouguang, Shandong Province, the experimental results show that compared with the traditional neural network model, the proposed VMD-CNN-LSTM model can effectively improve the precision of short-term load prediction of agricultural greenhouses.