Under the strong guidance of the new energy vehicles to the countryside policy, sales of electric vehicles in rural areas have grown rapidly. However, rural power grid is widely distributed, power supply lines are long, and charging load is relatively dispersed and difficult to predict. To this end, a charging load prediction model for electric vehicles in rural areas based on modified graph temporal convolutional network（MGTCN）is proposed. Firstly, a rural power grid graph structure matrix is constructed based on graph convolutional neural network to characterize the spatial information of user charging characteristics and reduce the dimension of input data. Secondly, a temporal convolutional network is introduced to perceive the time series information of charging data and mine the time series features that affect load forecasting. Then, an MGTCN algorithm based on attention mechanism is proposed for charging demand forecasting. The attention mechanism assigns different weights to each feature, and the model can adaptively learn network parameters. Finally, the effectiveness of proposed method in predicting electric vehicle charging load in rural areas is verified based on the example results, and the impact of charging load on rural power grids under different electric vehicle penetration rates is further analyzed.