Abstract: Considering the power grid fault， the diagnostic information redundancy is low based only on protective relays and circuit breakers(switch) for diagnosis， and the accuracy of diagnosis will be affected under complex fault cases. The electric data information was introduced to propose the model prediction and data cleaning method， as well as to establish the power network fault diagnosis system. By using the model to predict the electric quantity information accurately， the cleaning rules and the logical inference rules were established， and the data cleaning and verification of the switch were carried out respectively. On this basis， the abductive reasoning network (ARN) was used to diagnose the fault information， and the candidate faults were obtained. The simulation results verified the validity and accuracy of this method.

Key words: fault, diagnosis, data cleaning, model prediction, abductive reasoning network