Abstract: Complex geological conditions, long drilling cycle and difficult wellbore pressure control are common problems in deep formation oil and gas resource development. The intelligent well killing method, combined with multi-source real-time information feedback, can realize real-time prediction and update of gas-liquid distribution and pressure change law in the wellbore. However, the combination of different correction factors involved in the intelligent kill model may get the same pressure calculation result, which leads to the problem of multiple solutions in the model. By analyzing the evolution law of the spatial form of the solution at different historical time nodes, it is revealed that the essence of the multi-solution of the model comes from the imperfection of the model training under the constraint of a small amount of data. The model global training optimization method based on real-time information sequence and the dynamic random population training optimization method are established correspondingly, and their search ability and applicable conditions for the global optimal solution of the model are tested. The results show that the global training optimization method can achieve accurate control in the early stage of well killing, but the calculation time is long. The dynamic random population training optimization method is slightly different from the expected value in the initial stage of well killing, but the calculation time is less. According to the available computing resources, a suitable training optimization method can be selected to realize the deep learning of the gas-liquid flow law under the constraints of multi-source real-time data.