Fault arc in photovoltaic power generation system is difficult to accurately detect due to strong randomness, weak signal, and easy to be affected by load sudden change. According to the U-I output characteristics of photovoltaic cells, this paper analyzes the generation mechanism of DC series weak fault arc in photovoltaic power generation system, and analyzes the characteristics of weak DC series fault arc signal by building a photovoltaic power generation system fault arc simulation experimental platform, and then a method to detect weak DC series fault arc based on the wavelet energy entropy features of current signal is proposed. The proposed method firstly calculates the current pulse factor, which are used to detect the fault arc with threshold comparison method. On this basis, the current wavelet energy entropy features are calculated to identify weak fault arc based on extreme learning machine (ELM). The experimental results show that the proposed method can not only detect strong DC fault arc, but also detect weak DC fault arc with high average identification rate 98% .