【作者】 高明；

【导师】 楚纪正；

【作者基本信息】 北京化工大学 ， 控制科学与工程， 2013， 硕士

【摘要】 石油化工行业是支撑我国国民经济重要产业。随着现代化工生产过程越来越复杂，原材料日益短缺，能源价格不断上涨，化工行业竞争越来越激烈，提高化工过程生产的效率，增加企业的经济效益，以及减少原材料和能源的消耗已经成为迫切要解决的问题，因此对化工过程优化的研究越来越受到重视。而化工过程机理复杂，建立化工过程对象的机理模型是困难的、耗时的，因此采用基于生产数据的经验建模方法是符合实际的。本文通过分析化工过程优化的重要性，开发了一套以化工过程为对象，立足生产数据建立经验模型，以试验设计方法为优化算法的数据驱动优化软件。本文主要工作及研究内容如下：(1)提出了运用高斯过程回归代替人工神经网络和支持向量回归建立对象模型，进而改进了Chen等人提出的基于人工神经网络（ANN）和信息分析法的过程优化算法，调整了优化算法中下批次数据点的选取方法，将优化结果点中性能最好的一组作为下一批次试验数据点。以常压塔为优化对象，常压塔全塔经济效益为性能指标，对改进的算法进行了验证，结果显示改进的算法使得常压塔获得了更好的经济效益，验证了改进算法的有效性。(2)提出并设计了以生产数据为操作源的数据驱动优化软件。软件系统的整体架构分为表现层，功能层和数据层，软件系统主要包括数据处理、模型选择和优化操作三大功能模块，根据三层结构和三大功能模块，实现了数据驱动优化软件。(3)将本文设计和实现的优化软件应用到常压塔的操作优化，分别用人工神经网络、支持向量回归和高斯过程回归建立常压塔模型，实现了常压塔优化，优化结果证明了该优化软件的有效性和可行性，满足了化工过程操作优化功能。更多还原

【Abstract】 Petrochemical industry is important industry which has sustained nationaleconomy. Due to the complexities of the modern chemical processes, rawmaterial shortages, increasing energy prices, and the strong competitionsbetween the chemical industries, improving the efficiencies of chemicalprocess production, improving the economic benefits of the enterprises anddecreasing the consumptions of raw materials and energy sources has alreadybeen a compelling problem. So the studies of chemical process optimizationhave been paid more and more attention to. However, because of thecomplexities of mechanism of chemical processes, building the mechanismmodel of chemical process is difficult and time-consuming. Empirical modelsthat based on production datas are efficient. This paper analyses theimportances of chemical process optimization and develops a data drivenoptimization software that uses chemical process as objects. We buildempirical models by the production datas and use the design of experiments asoptimization algorithm.In this paper, the main work and research contents are listed as follows:(1) Aiming at Chen proposed a design of experiment optimization method that based on artificial neural network(ANN) and information analysis(AIDOE), the article improves it, which to use a gaussian process regression(GPR) to build the model of the objects, instead of artificial neural networkand supporting vector regression(SVR). At the same time we adjust the choiceof the data points in the optimization algorithm, which to choose the best oneof optimization results as the next experimental data points. The atmosphericdistillation unit (ADU) is used as the optimization object. The improvedalgorithm can get the best economic benefits of the ADU. It demonstrates thatthe improved algorithm is effective.(2) It puts forward and designs a data driven optimization software.Software system architecture consists of display layer, function layer and datalayer. Also, it mainly contains three modules: data processing, modeling andoptimizing. According to the architectures and modules, we achieve theoptimization software.(3) The data driven optimization software is applied to the optimizationof ADC. The model is builded by ANN、SVR and GPR. The availability andfeasibility of my software are demonstrated by the optimization results. And itsatisfies the function of the chemical process optimization.更多还原