封面图片
2023年, 第17卷 第10期
COVER
Solving the integrated problem of long-term planning and short-term scheduling in a large-scale refinery-petrochemical complex has been a big challenge over the past years. The mixed-integer nonlinear programming model is usually intractable due to the vast binary variables which increase the computational expenses. Therefore, a data-driven approach is proposed based on the idea of predicting the final objective with given integer variables. A convolutional neural network (CNN) is implemented for such purpose and combined with classical optimization as a heuristic algorithm. In this approach, the CNN is first used to decide whether the given integer variables are the potential to provide a satisfactory solution. If the prediction is positive, then the resulting nonlinear model is solved to optimality. The final solution is updated when a better solution is obtained. Via such a framework, the complex integrated planning and scheduling problem can be solved much more efficiently compared to commercial solvers BARON and SCIP, showing advantages in a realistic application. (Lifeng Zhang, Haoyang Hu, Zhiquan Wang, Zhihong Yuan, Bingzhen Chen, pp. 1516–1532) [展开] ...
ISSN 2095-0179 (Print)
ISSN 2095-0187 (Online)
CN 11-5981/TQ
Postal Subscription Code 80-969 Formerly Known as Frontiers of Chemical Science and Engineering in China 2018 Impact Factor: 2.809