To solve the fault prediction of partially observed discrete event systems (DES), this paper proposes a prediction method for the fault probability and occurrence date of DES based on labeled time Petri nets (LTPN). By using the modified state class graph of an LTPN system, the feasible paths consistent with an observed time label sequence are computed so as to establish the preliminary fault diagnosis results. The occurrence probability of each detected fault class is calculated based on its probability density distribution and the Gauss-Kronrod integration method. Subsequently, the occurrence date of each fault class is predicted, enabling proactive prevention of faults. Finally, the availability of the proposed method is demonstrated through an alternating bit protocol. The results show that the proposed method can effectively estimate the fault probability and provide targeted information on fault occurrence date. In particular, its application is expected to enhance the efficiency and accuracy of fault diagnosis in practical systems, and prevent the occurrence of faults in advance, reducing unnecessary loss caused by system failures.