The detailed scheduling of distributing products through pipelines is one of the most essential tasks in the management of multiproduct pipelines. The current methods of scheduling optimization are still limited by the dilemma between optimality and efficiency, especially when dealing with real-world scheduling issues. This paper introduces a two-stage mathematical model based on the event sequence, whose model scale greatly decreases after stripping the terms of event sequence. Then, a data-driven method is presented to learn a large number of existing scheduling data and accelerate the calculation of first-stage model. The method is decomposed into three parts: (1) take cold start to generate a good deal of basic training data; (2) train the neural network for the fast solution of first-stage model; (3) take real-world cases to improve the self-learning of neural network. A multiproduct pipeline in China is taken as the example to prove the practicability and superiority of the proposed method. The experimental results show that the proposed method could decrease the computation time from about several hours to several minutes. (C) 2019 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.