Improved MINLP Algorithms And Its Application In Production Scheduling Of Ethylene Process

The mathematical models for production scheduling in petroleum chemical processes are usually mixed integer nonlinear programming (MINLP) models, which contain many discrete and continuous variables and nonlinear functions. The characteristics of discontinuity and many local optimums in feasible search space present a big challenge in the searching efficiency and optimality guarantee for existing mathematical programming algorithms. Therefore, the algorithms for MINLP have been the one of the hot issues and difficult tasks for the research community.This thesis focuses on the design and implementation of improvement strategies for MINLP algorithms. The efficiency of the improved algorithms is illustrated based on the Benchmark problems, and then these algorithms are applied to solve large-size scheduling problems in practical ethylene production. Based on the proposed mathematical models and algorithms, the corresponding decision support system for the ethylene production process is developed.1. According to the structural characteristics of MINLP models, and taking into consideration the limitations of existing MINLP algorithms, four kinds of improved MINLP algorithms are presented.1) A hybrid outer approximation (OA) algorithm is developed by incorporating the generalized Benders decomposition (GBD) into the OA algorithm;2) A parallel GBD algorithm is presented by solving the sub-problems of GBD in a parallel way;3) A hybrid parallel OA algorithm is designed by incorporating the GBD and at the same time solving sub-problems in a parallel way;4) A quadratic OA algorithm is proposed by taking the mixed integer quadratic programming as an approximation of the original problem when solving the master problem.2. A flexible algorithm platform for MINLP is built to implement the four improved algorithms. The results of numerical experiments based on Benchmark cases demonstrate the correctness, reliability and efficiency of the algorithm platform. The four improved algorithms are tested on this platform using Benchmark problems, and the obtained results are compared with those obtained by DICOPT, which is a standard software package of OA algorithm. The comparison shows that significant improvements can be obtained by the proposed four algorithms in the iterations and computational times that are needed to reach the optimal solutions.3. According to the characteristics of domestic ethylene production processes, two production scheduling models are formulated and then solved using the presented improved MINLP algorithms. The computational results show that the solving efficiency can be significantly improved by the four improved algorithms for practical production scheduling problems of the ethylene process and consequently the improved algorithms hold an advantage in the CPU time needed to obtain optimal solutions. An optimization system for the production management of ethylene cracking furnace is developed based on the practical production management of a certain domestic ethylene plant. The implementation of this system can provide not only a convenient tool for the management of production data, but also the optimal solutions of ethylene production for decision-makers. In addition, a human-computer interaction module is also incorporated into this system to further modify or improve the obtained schedules.Based on the obtained computational results, it is illustrated that the improved algorithms for MINLP proposed in this thesis can significantly improve the solving efficiency for the practical scheduling of ethylene production, and therefore it can concluded that the improved algorithms have potential applications for practical MINLP problems. The developed optimization system for the production management of ethylene cracking furnace integrates the information and optimization technologies, and provides a visualization platform with a human-computer interaction module to obtain optimal production schedules in a scientific way.