Fault location methods for single-phase-to-earth in small current neutral grounding system are difficult to meet the needs of the site, which is too simple to have a high accuracy in actual run-time. This paper presents a fault location method of multi-information fusion based on extreme learning machine. For real-time measured zero-sequence transient current, the terminal uses transient energy method, wavelet method, and the first half-wave method to extract feature vector and uploads it to the host station; inputs to the extreme learning machine network which has the be-trained parameters. Host station restarts multi-information fusion fault location and output search result. The method is less affected by ground position, grounding moment, ground transition resistance and other factor, is applicable to different single-phase ground case, which has higher fault location robustness. The on-site experiments are carried out to prove the feasibility of the method.