To ensure gas supply reliability and reduce the risk of gas shortage for customers, a deep learning-based method for the optimization of peak shaving of natural gas pipeline was proposed. Firstly, a forecasting model of node pressure based on deep learning was developed, aiming to capture the impact of boundary transient changes on the node pressure. Then, a multi-objective optimization model with minimum operation cost and maximum pipeline gas storage capacity was established, which could dynamically optimize the gas storage and gas compressor operating scenarios. Finally, the proposed method was validated on a natural gas pipeline network. The results show that, compared with the operation scenario without optimization, the average gas storage capacity of the pipeline increased to 33.628 million cubic meters after optimization, and the operating cost was saved by RMB 106 000. The results can provide a new insight into the peak shaving of natural gas pipeline systems.