CIESC Journal ›› 2016, Vol. 67 ›› Issue (12): 5169-5175.DOI: 10.11949/j.issn.0438-1157.20161287

Previous Articles     Next Articles

Parallel algorithm of hybrid BB/SQP for heat exchanger network synthesis

JIANG Nan1, LIU Yongzhong1,2, ZHU Tianhong1   

  1. 1 School of Chemical Engineering and Technology, Xi'an Jiaotong University, Xi'an 710049, Shaanxi, China;
    2 Key Laboratory of Thermo-Fluid Science and Engineering, Ministry of Education, Xi'an 710049, Shaanxi, China
  • Received:2016-09-13 Revised:2016-09-22 Online:2016-12-05 Published:2016-12-05
  • Supported by:

    supported by the National Natural Science Foundation of China(21376188, 21676211).

换热网络合成问题的并行BB/SQP混合算法

姜楠1, 刘永忠1,2, 朱天鸿1   

  1. 1 西安交通大学化学工程与技术学院, 陕西 西安 710049;
    2 热流科学与工程教育部重点实验室, 陕西 西安 710049
  • 通讯作者: 刘永忠。yzliu@mail.xjtu.edu.cn
  • 基金资助:

    国家自然科学基金项目(21376188,21676211)。

Abstract:

Heat exchanger network synthesis can be described by a mixed integer non-linear programming(MINLP) model, which features non-convex, non-linear and non-differentiable optimization. The parallel computing technology based on GPU provides an efficient support for solving large scale models. In this work, a hybrid algorithm combined branch and bound method(BB) with sequential quadratic programming(SQP) is proposed to overcome difficulties in the existing parallel SQP algorithm, such as too many combinations of integer variables, dependency of initial values and etc.. The BB method is adopted in the hybrid algorithm instead of the exhaustive method. It can not only reduce the combinations of integer variables, but also select feasible initial values for the SQP algorithm. The solution quality is improved. The results of the examples show that the proposed parallel hybrid algorithm can solve the heat exchanger network synthesis problems efficiently. Compared to the serial algorithm, the proposed parallel algorithm has much higher executive speed with the speedup ratio of 39.

Key words: heat exchanger network synthesis, mixed integer non-linear programming, GPU, hybrid algorithm

摘要:

换热网络合成问题通常可用非凸、非线性、不可微的混合整数非线性规划模型描述。基于GPU的并行计算技术为求解大规模模型提供了高效支撑。针对已有并行SQP算法求解换热网络合成问题中存在二元变量组合数过多、并行SQP算法求解结果严重依赖初值等问题,提出了BB/SQP混合并行算法。该算法采用BB算法代替枚举法,不但大大减少了模型求解中可能的二元变量组合,而且为SQP算法选出了可行的初值,从而提高了算法的求解质量。研究表明,所提出的混合并行算法能够有效求解换热网络合成问题,且并行计算相比串行计算的求解速度显著增加,加速比可达39。

关键词: 换热网络合成, 混合整数非线性规划, GPU, 并行算法

CLC Number: