Abstract: For the shortcoming that typical fuzzy operators are lack of flexibility, generalized probability sum (GPS) fuzzy operator and generalized probability product (GPP) fuzzy operator are introducted into fuzzy neural network (FNN). The transfer functions of neurons in rules and output layer are replaced by GPS and GPP, and the strength of logic operation are adjusted using compensation parameters to simulate the flexibility of human thinking. A model based on GPS-GPP FNN for fault prognosis is proposed, the parameter of iterative algorithm is derived for training. The model is simulated track circuit fault prognosis system, and a mapping relationship based on cotangent function between reliability and maintenance date limitation is proposed. Through comparing the prognosis results of GPS-GPP and Sum-Prod., it is proved that the GPS-GPP fuzzy neural model has better accuracy and generalization ability.