The invention discloses a transient stability real-time assessment method and device based on a deep belief network. The method comprises the steps that a learning sample set is generated by using a time-domain simulation technology; system measurement is taken as model input, a stable state is taken as a model output, the network parameter of the deep belief network is updated through unsupervised pre-training and supervised precise adjustment, and a transient stability assessment model is formed; the data actually measured by a fault clearing moment system is input into the transient stability assessment model, and the transient stability of the system is predicted. According to the method, the electric power system characteristics can be automatically extracted by using the DBN so as to be used for the transient stability assessment, and the requirements for the calculation speed and accuracy of real-time assessment of the transient stability can be met at the same time, the transient stability real-time assessment can be achieved, the assessment efficiency is improved, the assessment accuracy is improved, and the method is simple and easy to implement.