【作者】 郑小青；

【导师】 邵之江；

【作者基本信息】 浙江大学 ， 系统工程， 2006， 硕士

【摘要】 在流程工业自动化系统中,优化起着非常重要的作用,如何提高优化的求解速度和精度是一个意义重大且富有挑战性的研究课题。而在优化计算过程中,导数是至关重要的信息,因为它代表了寻优过程的搜索方向。同时,导数求解的耗时在整个优化时间中往往占有很大的比重,尤其是在对大规模复杂计算流程进行优化时,求导耗时往往成为提高优化效率的瓶颈。由此,寻找一种精确又快速的求导算法是提高优化计算效率最有效的途径,这也是本文工作的重点所在。 通过研究当前热门的自动微分求导算法,并将其应用于精馏塔灵敏度分析过程,以及应用于本文建立的乙烯生产流程脱丙烷和脱丁烷塔的联塔模型的优化计算过程,再与传统的差分法和稀疏差分法进行比较,分析求导算法的不同对优化效率的影响。 本文的研究工作包括以下几个方面: 1.简要介绍过程系统优化的重要性,通过分析求导和优化的关系,强调求导在优化计算中的重要性。并介绍自动微分算法的发展和应用、基本原理和自动微分算法的发展。再着重分析和比较自动微分算法的两种实现方式:基于操作符重载和基于编译原理,并同传统的差分法和稀疏差分法进行比较分析,找寻最可行有效的求导算法。 2.对乙烯生产流程中两个重要的单元脱丙烷塔和脱丁烷塔,采用开放方程法在MATLAB平台建立联塔数学模型,并对联塔模型进行模拟计算,将计算结果同ASPEN流程模拟结果进行比较,以验证模型的合理性。 3.实例分析自动微分算法在精馏塔优化计算的灵敏度分析中的应用,包括将其应用于灵敏度分析过程,以及灵敏度的并行计算过程。 4.实例分析自动微分算法在过程系统优化计算中的应用,包括将自动微分应用于乙烯生产流程的脱丁烷单塔以及联塔的优化计算中,并比较和分析其对优化效率的影响。更多还原

【Abstract】 Process systems optimization plays a significant role in the field of Process Systems Engineering. And how to improve the efficiency and accuracy of optimization is a big challenge for today’s researchers. Since the numerical derivative represents the search direction of optimization process, it’s the key information for optimization. Besides, derivative calculation is always the most time-consuming part of the whole optimization process, especially for the large-scale problems. Therefore, applying an accurate and efficient derivative calculation way to optimization is of great importance for the improvement of optimization efficiency.Automatic differentiation, a promising differentiation algorithm, which has been developed rapidly in recent twenty years, is investigated in the thesis. And automatic differentiation is applied to the sensitivity analysis problem of distillation column optimization, as well as the optimization problem of the multi-distillation columns model of depropanizer and debutanizer in ethylene plant. Finally the calculation results and efficiency are analyzed and compared with the traditional finite difference and sparse finite difference.The main contributions include the following aspects:1. The importance of numerical derivative for optimization is studied by analyzing the relationship of derivative evaluation and optimization. And the development and applications of automatic differentiation algorithm in recent years are introduced, as well as the basic algorithm is discussed. And two implementation ways of automatic differentiation based on source transformation and operator overloading are fully investigated and compared with the traditional methods of finite difference and sparse finite difference.2. Equation oriented method is applied to the multi-columns modeling of depropanizer and debutanizer in ethylene plant based on first principles. And the simulation result of the built model is proved correct after comparing with the result of Aspen Plus, a professional and commercial simulation software.3. The application of automatic differentiation in sensitivity analysis of distillation column optimization is discussed, including applying it to thesensitivity analysis problem of distillation column optimization, as well as the corresponding parallel sensitivity analysis process.4. The application of automatic differentiation in process systems optimization is fully investigated, such as applying it to the optimization problem of the multi-distillation columns model of depropanizer and debutanizer in ethylene plant, as well as the optimization problem of single debutanizer model. The influence of different derivative evaluation ways to the efficient of optimization is analyzed and compared through the applications.更多还原