换热网络合成问题的并行BB/SQP混合算法  

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

作　　者：姜楠[1] 刘永忠[1,2] 朱天鸿[1] 

机构地区：[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

出　　处：《化工学报》2016年第12期5169-5175,共7页CIESC Journal

基　　金：国家自然科学基金项目(21376188;21676211)~~

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。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.

关 键 词：换热网络合成 混合整数非线性规划 GPU 并行算法 

分 类 号：TQ021.8[化学工程]