Spatial load forecasting is of great significance for distribution network planning and construction. In order to improve the accuracy of distribution network spatial load forecasting, a distribution network spatial load forecasting method based on entropy weight method and grey relational analysis-extreme learning machine (GRA-ELM) is proposed. Firstly, the districts in the planning area are seperated according to the nature of land use. The influencing factors of different types of load are analyzed, and the index system of spatial load density is established. Secondly, the entropy weight method is used to distribute the weight of load density index for different types of load. Then, GRA is used to select the training samples which are similar to the load density index of the plot to be tested. Finally, the samples are brought into the extreme learning machine (ELM) model after the parameters are optimized by particle swarm optimization (PSO), and the prediction results are obtained. The performance of the proposed method is verified by an example and the results show that the proposed spatial load forecasting method has high accuracy than other methods does.