The invention is applicable to the technical field of nuclear power, and provides a method, a device and a terminal device for predicting the change trend of the running state of a generator. The method is applied to a megawatt-level pressurized water reactor nuclear power plant. The method comprises the following steps: an inference engine is triggered to run when online real-time data trigger aprecursor, and the fault is diagnosed to obtain a first diagnosis result; when the manually submitted non-real-time precursory information is received, the inference engine is triggered to run, the fault is diagnosed, and the second diagnostic result is obtained, wherein, the non-real-time precursor information comprises on-line non-real-time precursor information and test precursor information, and the test precursor information comprises abnormal symptom information occurring when the generator is tested in the process of overhauling; the first diagnostic result and/or the second diagnosticresult are analyzed to obtain trend information indicative of a change trend of a generator operating state. The invention can obtain the trend information characterizing the change of the running state of the generator in advance, and has strong practicability and ease of use.