There is a problem of low sensitivity and difficulty in identifying high-resistance ground faults in existing fault detection methods for ultra high voltage direct current (UHVDC) transmission system. Thus a fault identification method for a UHVDC transmission system based on the integer factor (IF)-approximate derivative (AD) and an extreme learning machine (ELM) is proposed. The IF is used to analyze the signals at different sampling frequencies, and the AD method is used to obtain different degrees of detail coefficients for the signals. First, the signal is down-sampled based on different IFs, and the AD method is used to calculate the first, second and third order approximate derivatives of the obtained signal. Secondly, the entropy characteristics of each sub-signal are calculated. Then, recursive feature elimination with a cross validation (RFECV) algorithm is used to screen the features of the obtained series of features, and the ELM is used to identify the UHVDC transmission system fault types. Finally, the UHVDC system model of ±800 kV is built in the Matlab/Simulink environment to simulate different fault types. The experimental results show that the proposed method has higher accuracy and strong tolerance to transition resistance when identifying different types of faults in a UHVDC transmission system.