Abstract: The incorporation of progressive collapse design is crucial to improve the resistance of structures against accidental local loads. Among current design codes, the alternative path (AP) method is often recommended. In which, the nonlinear dynamic AP method first simulates the structural responses after the component removal, then the reinforcement within the connected components is adjusted based on the simulated responses and the code requirements to ensure the safety of the structure. Compared with other AP methods, the nonlinear dynamic AP method is accurate in calculation but requires a large number of nonlinear dynamic AP analyses. In recent years, deep learning has been widely used in solving engineering problems. By learning the inherent laws of sample data, deep learning can extract the structural features and predict the structural response under component removal scenarios. In this study, a hybrid design algorithm for the progressive collapse design of reinforced concrete (RC) frame structures was proposed by combining deep learning and particle swarm optimization. The reinforcement feature map was constructed to depict the structural progressive collapse resistance. Machine learning and deep learning were respectively employed to develop the prediction models to estimate the structural progressive collapse responses based on the feature maps. The particle swarm optimization (PSO) algorithm was combined with the prediction models to optimize the progressive collapse design of RC frame. On this basis, a 4-story RC frame structure was taken as an example, and the design results were compared with those calculated based on the finite element (FE) method to verify the feasibility and accuracy of the hybrid design algorithm.