Spatial load forecasting is of great significance to the planning and construction of distribution network. In order to improve the accuracy of spatial load forecasting of distribution network, based on extreme learning machine, a spatial load forecasting algorithm is put forward in this paper. The parameters of the particle swarm optimization mode are adopted. Firstly, the load is classified according to the property of land use. Then, the FCM algorithm is used to carry out cluster analysis for each type of load and a refined load density index system is established. Next, the training samples are carried out with the extreme learning machine to improve the accuracy of prediction, which selected according to the characteristic indexes of the plots to be predicted. The example is simulated by the search data. By comparing with the relative error without the introduction of FCM algorithm, the relative error without the introduction of PSO optimization algorithm and the relative error with the adoption of PSO-ELM algorithm, it can be obtained that the PSO-ELM algorithm proposed in this paper has a high accuracy and meets the requirements of practical engineering.