摘要
针对甲醇烃动态过程反应参数优化问题,提出了一种基于混合自动微分技术和改进简约空间序列二次规划(rSQP)算法相结合的求解方法.该方法将动态优化问题离散化为以代数方程表示的非线性规划问题,利用问题结构稀疏、自由度相对较低,并含有大量等式约束等特点,以改进的简约空间序列二次规划算法为求解器来求解优化问题,并在求解过程中,采用混合自动微分技术获取优化问题的一阶导数信息和稀疏结构.计算结果表明,该方法的求解效率比差分求导的标准序列二次规划(SQP)算法高100多倍,比混合自动微分求导的SQP算法高10倍左右,另外求解精度也有显著提高.
To optimize the reaction parameters in the dynamical process of methanol-to-hydrocarbons, a method based on reduced sequential quadratic programming (rSQP) and hybrid automatic differentiation technology was presented. The dynamical optimization problem was firstly discredited as a nonlinear programming problem denoted by algebraic equations. With the characteristics of sparseness, relatively low degrees of freedom and equality constraints utilized, the nonlinear programming problem was solved by improved rSQP solver. In the solving process, hybrid automatic differentiation technology was used to obtain the sparse structure and gradient information. Computational results show that the efficiency of the proposed method is more than 100 times as that of the standard sequential quadratic programming (SQP) with differences, and is more than 10 times as that of standard SQP with hybrid automatic differentiation. The accuracy is also improved with the proposed method.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2004年第12期1606-1610,1614,共6页
Journal of Zhejiang University：Engineering Science
基金
国家"863"高技术研究发展计划资助项目(2002AA412110)
国家"973"重点基础研究发展规划资助项目(2002CB312200).
关键词
过程系统优化
简约SQP
自由度
梯度
自动微分
Computational methods
Degrees of freedom (mechanics)
Differential equations
Gradient methods
Process control
Quadratic programming
