Natural gas pipelines have complex geographical environments and varied operating conditions. Uncertainty simulation represented by Monte Carlo methods has become the main method for pipeline corrosion reliability assessment. However,the high-order simulation problems caused by the high design reliability of natural gas pipelines make Monte Carlo simulations very time-consuming. In order to solve this problem, this paper uses a neural network algorithm rather than Monte Carlo simulation to establish a nonlinear model of basic pipeline parameters and reliability. Because of the difficulty of combining prior information in the modeling process, this paper proposes an innovative method that combines an intelligent optimization algorithm and a neural network algorithm. This method can incorporate the pipe corrosion reliability variation into the modeling process. An integrated computational flow from the selection of feature variables, the generation and processing of sample data,the construction of neural network models and the evaluation of model prediction effects are proposed. Under various working conditions, the neural network model constructed by the method proposed in this paper predicts the reliability of pipeline structures. The results show that the model can obtain the calculation results highly similar to Monte Carlo simulation in a very short time. Compared with the traditional neural network model, the model established by this method has greatly improved the reliability of prediction and the ability to reflect changes in reliability.