【作者】 缪啸华；

【导师】 王建华；

【作者基本信息】 华东理工大学 ， 控制科学与工程， 2012， 硕士

【摘要】 本文依托于某公司甲醇生产过程优化控制项目,主要研究该项目中甲醇合成塔出口的粗甲醇质量转化率的软测量问题,将模糊神经网络和差分进化算法引入软测量建模,采用改进的差分进化算法对甲醇生产中转化率的测量问题进行了研究并着重分析了其在合成工艺中的应用。本文以甲醇合成为研究对象,针对系统软测量模块问题进行了深入的研究分析,探索了几种有效的能够解决甲醇实际生产过程中存在的质量参数的软测量问题。首先,总体介绍了软测量系统结构及相关建模方法,介绍了甲醇合成工艺流程,通过对甲醇合成过程工艺规律及动力学的研究,明确了甲醇合成过程的重要质量参数,并确立了研究对象和目标。然后,通过对几种典型建模方法的研究,提出了基于模糊神经网络的建模方法,并对差分进化算法进行深入研究,提出了改进差分进化算法,该算法加入了单纯形寻优操作和自适应变异算子和时变交叉概率操作,根据寻优过程中的信息,智能的调节算法中参数的设置,避免在求解过程中陷入局部最优,并提高全局寻优能力和收敛速度,以解决标准差分进化算法所存在的收敛速度慢,以及系统资源开销较大等问题。其次,将本文提出的改进型软测量方法用于实时分析预测甲醇合成塔出口粗甲醇的质量转化率,并通过比较标准差分进化算法及改进差分进化方法的优缺点,表明本文所提出的改进差分算法能够更加快速、有效地逼近预期结果,提高预测精度。最后,采用Visual C++平台及组态王软件相结合,实现了甲醇生产过程软测量系统的开发。本文所研究的软测量算法,在确保系统的实时性和可靠性的同时,能够快速、准确、有效的实现对观测量的预测估计,从而极大的节约了测量成本,提高生产效益。更多还原

【Abstract】 This thesis is sponsored by a company process optimization control of methanol project. Focus the mainly studies on the soft sensor of the methanol conversion rate. Some researchers have been done on the soft sensor of conversion rate in conversion process of methanol production. Improved differential evolution algorithm is applied in fuzzy neural network model soft sensor.Based on research platform of the methanol project, generate researches have been done in soft sensor area to find effective schemes to solve the soft sensor problem.Firstly, the architecture of soft sensor system and its key technologies are reviewed by studying some typical model methods, and the improved differential evolution algorithm has been proposed and used in soft sensor. In order to avoid getting into local optimum and improve the global optimization ability, a improved algorithm based on differential evolution and simplex method is proposed to solve and improve the algorithm convergence speed and global optimization efficiency, as well as to avoid the high system resource cost.Secondly, the algorithm is used for the prediction of conversion rate in conversion process of methanol production. The comparisons with a differential evolution and improved differential evolution show that the soft sensor, which is proposed in the thesis, can approach the expected result effectively, and can also improve the prediction precision.Finally, C program language is used to develop a soft sensor system which is implemented in methanol project.The algorithm proposed in this thesis can realize the prediction more reliability, accurately and effectively, under the preconditions of guaranteeing real-time ability and reliable of system operation. Besides, the success of the project the system not only can save the survey cost, but also can increase the economic efficiency of enterprises.更多还原