Abstract: Under the background of accelerating transformation of China's energy structure and striving to achieve the goal of "dual carbon", the traditional power system will also usher in structural transformation. Due to the randomness, uncertainty and low inertia of renewable energy, a series of influences brought by large-scale new energy grid make the frequency safety problem of power system increasingly prominent. However, the traditional time domain simulation method has some disadvantages such as large amount of computation and long calculation time, so it is difficult to realize the rapid evaluation of the actual power system under the flexible and changeable operation mode and a large amount of measured data. In order to quickly assess the frequency security of power systems, a power system transient frequency security integrated assessment method based on deep residual shrinkage network (DRSN) is proposed. Deep residual shrinkage network introduces attention mechanism based on deep residual network, which can enhance useful information and suppress redundant information. On this basis, the samples were divided according to the maximum frequency change rate, and DRSN network was used for training respectively. Simulation results on the IEEE 39-bus system and IEEE 118-bus system added with wind power machines show that the proposed method has higher accuracy and excellent generalization、robustness and applicability.