文档介绍：∗
基于多GPU的换热网络MINLP模型求解优化
夏明星 1,任宇星 1, 唐亚哲 1, 康丽霞 2, 刘永忠 2
1(西安交通大学计算机科学与技术系,陕西西安 710049)
2(西安交通大学化工系,陕西西安 710049)
Optimization on the Solution of Heat work MINLP Problems Based
on Multi-GPU*
XIA Ming-Xing1, REN Yu-Xing1, TANG Ya-Zhe1, KANG Li-Xia2, LIU Yong-Zhong2
1(Department puter Science and Technology, Xi’an Jiaotong University, Xi’an 710049, Shaanxi,China)
2(Department of Chemical Engineering, Xi’an Jiaotong University, Xi’an 710049, Shaanxi, China)
Abstract: This paper presents a puting framework, a number puting tasks scheduling algorithms
and the corresponding parallel algorithms for screening of optimal solutions for solving the MINLP problem. We
concentrate on both puting model and specific GPU-CUDA programming level optimization. Tests on a
simple MINLP problem and two small-sized heat work models are conducted. The results show the
new algorithm has better optimal solution and faster execution than the one running serially on CPU.
Key words: GPU; CUDA; MINLP; heat work
摘要: 本文设计并实现了基于统一计算架构(CUDA) 的并行求解混合整数非线性规划模型的框架以及计
算任务的分配调度算法。针对不同的计算任务调度算法设计并实现了相应的并行筛选最优解算法,并在并行
模型和算法细节实现上采用了各种并行优化方法。对一个简单 MINLP 模型和两个实际换热网络实例的测试
结果表明,本文模型与串行算法相比,结果更优且加速比明显。
关键词: GPU;CUDA;MINLP;换热网络
中图法分类号: TP301 文献标识码: A
换热网络是炼油、化工等高耗能工业中能量高效利用的关键子系统之一,可抽象成为一种混合整数非线性
规划(mixed integer non-linear programming, MINLP)问题[1][2][3]。对于换热网络的求解的研究较多的集中在数学
规