Abstract: In order to solve the space registration problems of multi-sensor in the photoelectric theodolite measurement, a space registration model based on the extreme learning machine(ELM) algorithm was proposed in this paper. Firstly, the ELM theory and the modeling steps of ELM space registration model were introduced. Then, the star measurement data was used to build ELM space registration model. Finally, the ELM space registration model was compared with single error correction model and spherical harmonics correction model. Experimental results indicate that ELM space registration method can improve the measuring precision of photoelectrical theodolite from about 17 to less than 1; the accuracy of the ELM space registration model is improved by more than 35% than single error correction model and spherical harmonics correction model. The results indicate that compare with the single error correction model and spherical harmonics correction model, space registration model based on ELM algorithm has higher prediction accuracy and stronger generalization capability.