The normal fault source location algorithm depends too much on the prior knowledge of distributed information systems, therefore, this paper presents a fault source location algorithm based on the invariant network and fault injection. Firstly, the invariant network model of the system is established by using the running data of the system components collected in the system log and in order to obtain the fault network model and form a set, the fault injection and diffusion modeling of each component are carried out on the model. Then, according to the value of the original time series, the data quality evaluation rules are formulated to judge whether the data has changed violently. Finally, the source of the system fault is determined using the method of local fitting between the actual fault network and the centralized fault network, the data quality evaluation rules are used as result revise to realize the accurate location of system fault source. The results on the synthetic data set, the data set collected by an open source distributed information system and the actual data set of a power grid dispatching system show that the proposed method has higher accuracy.