This work aims to report a numerical paradigm for identifying plastic hinges of inelastic nonlinear steel space frames with semi-rigid connections. Data randomly created by the advanced analysis method is used to build deep learning (DL) models. The input data are displacement at several important degrees of freedom (DOFs), whilst the output ones are the position and yield ratio of semi-rigid connections. The geometric nonlinearity, i.e. both N-δ and N-Δ effects, is taken into account by the stability functions. Column Research Council (CRC) tangent modulus and Orbison yield surface are employed to model the material nonlinearity. Semi-rigid connections are simulated by three-parameter models suggested by Chen and Kish. Accordingly, the information about the plastic hinges including the position and yield ratio can be monitored easily. A six-story space steel frame is tested to illustrate the ability of the proposed method. Obtained results have shown that Deep Neural Network (DNN) can reliability diagnose plastic hinges that occur in nonlinear steel frames. All code structures are programmed by Python with version 3.7 on a laptop computer.