【作者】 安娜；

【导师】 潘立登； 陈标华； 李成岳；

【作者基本信息】 北京化工大学 ， 控制理论与控制工程， 2003， 硕士

【摘要】 近年来,环境问题引起了世界各国的关注。作为空气污染的主要来源之一,易挥发有机物(VOCS)不仅是一次污染源而且还可能造成光化学烟雾等二次污染。因此,控制大气中挥发性有机物非常重要。在这一方面,周期性改变流向的非定态催化燃烧技术——流向变换催化燃烧技术具有独特的优势。流向变换催化燃烧反应器将催化燃烧和反应热回收过程高度集成,能够大大节约投资和操作费用。随着计算机应用、自动控制技术的不断发展,对流向变换催化燃烧过程实行先进控制和优化操作,推动这一技术工业化和拓展其应用领域,是当前十分迫切和重要的工作。事实上,工业废气中VOCS的组成、浓度甚至气量都有可能频繁波动,因此,在这些输入参数波动的情况下,通过适当调节换向周期、空气流量等控制变量,使反应器正常操作,即既不“飞温”,也不“熄火”,而且具有很高的VOCS转化率,是实行先进控制的最终目标。热波的形状与特征参数,如波峰温度、平均温度和移动速度,决定了给定催化反应系统的性能,也提供了大量关于反应、传热系统有用的信息,是保证流向变换催化燃烧反应器正常操作的重要指标,必须及时而准确地预测并有效地控制。本文针对一个小型中试装置设计并建立了VOCS流向变换催化燃烧反应器监控系统,对原有的自动化程度较低的人工数据采集和控制<WP=4>方式进行了改造;用改进的RBF神经网络(RBFNN)建立了流向变换催化燃烧反应器拟定态温度分布模型,从过程的内在机理出发,提出了增强神经网络训练样本数据集正交性和完备性的方法;运用基于改进的RBFNN与遗忘因子最小二乘法(RLS)相结合的动态系统在线辨识方法,建立了反应器床层瞬态温度分布的预测模型;通过开展一系列控制实验,设计、建立了床层温度闭环控制系统,采用换向周期控制反应段平均温度和峰值温度;以旁路空气流量作为控制手段,提出了一种基于改进的RBFNN的非线性内模控制策略,并采用RLS在线校正模型参数,实现床层温度的自适应控制。研究结果表明:[1]监控系统由工控机、相关的板卡和MCGS组态软件编制的软件构成。实际运行表明,该监控系统具有可靠性高、实时性能好以及智能化程度高的特点。[2]神经网络模型泛化性能较差,其性能在很大程度上受到所选择的训练数据集的限制。从基于过程机理模型的数值计算结果出发,结合中试装置的实时操作数据,建立拟定态床层温度的人工神经元网络深层知识库,并采用改进的RBF神经网络,能够增强神经网络模型的“外推能力”和“可信度”。仿真结果表明所建立的模型简单,能满足控制要求。[3]用动态RBF神经网络建立的反应器床层瞬态温度分布的预测模型能够实现实时预测和控制,并在线校正模型参数。仿真结果与中试装置现场数据的对照表明,所建立的模型精度高,能满足实时控制<WP=5>的要求。[4]换向周期是保持正常操作的重要调控手段,是一个既能有效控制反应段温度水平,因而能在相当宽的变化范围内调节反应总量,又很容易改变的控制参数。[5]基于改进的RBFNN的非线性内模控制策略充分利用神经网络的自学习及非线性逼近能力,建立流向变换反应器的动态模型和逆模型,非线性内模控制器由逆模型和带有一个调节参数的鲁棒滤波器组成,并采用递推最小二乘(RLS)在线校正模型参数,仿真结果验证了此控制方案的有效性。总之,本项目从应用基础层面上对流向变换强制周期操作的人为非定态反应技术进行了系统研究,为进一步研究开发反应温度的高级控制技术、实现这一先进技术的国产化奠定了科学基础。更多还原

【Abstract】 In Recent years, environment protection draws international attention. Volatile Organic Compounds (VOCS), as one of the major sources of air contaminations, can not only cause direct pollution, but also lead to secondary pollution, such as photochemistry smog. Thus, it is very important to control the VOCS in the air. In this area, the reverse flow catalytic combustion technique is a unique solution. The reversed flow catalytic combustion reactor highly integrates the catalytic combustion and the recovery of the reaction heat, which cut down the operation cost and save the investment. With the developing of the control theory and the computer application technique, it is highly demanded to apply advanced process control algorithm and optimal operation for the reversed-flow catalytic combustion process.The composition and flow rate of industrial waste gas may vary frequently. In this case, maintaining the reactor in normal operation (e.g. the reactor will not go into temperature runaway or extinguish state) and high VOCs conversion by adjusting the reversal cycle duration or gas flow is the ultimate goal of the advanced control application. The shape of the thermal wave and the characteristic parameters, such as the maximum of the catalyst bed temperature, the average temperature and the shifting speed, determine the performance of the catalytic reactor system, and also provide lots of information regarding the reaction and the heat flow. Thus, they are important guidelines to maintain the reversed flow reactor in normal operation and must be timely and accurately predicted and be affectively controlled. <WP=7>Aimed at a pilot scale reverse flow reactor for catalytic combustion of VOCS in contaminated air, the advanced monitoring system was designed and built, with rebuilding the former manual data-collections and control mode. The quasi-steady state model of temperature profile for the reversed flow reactor was developed by the improved RBFNN (Radial Basis Function Neural Networks). A valid approach was proposed to enhance the data integrity and orthogonality of the neural network’s training samples. The RBFNN and the regressive least square method (RLS) with forget factor were adopted to develop the dynamic model for predicting temperature profiles. With developing a series of control experiments, a feedback control strategy for the flow reversal reactor was examined. The variable that was chosen to control average and maximum bed temperatures was the cycle period. A nonlinear internal model control (NIMC) strategy based on RBFNN was proposed by adjusting the gas flow, and the neural networks was trained online using RLS Algorithm, in order to realize adaptive control. The main conclusion may be summarized as follows:The controlling part of the monitoring system was mainly composed of industrial controlling computer, cards and the software written with MCGS configuration software. The system has been used in experimental environment, and the application of the system shows that the monitoring system is high reliability, real-time and intelligent.The application performance of Neural-network model is highly constrained by the selected training samples, it lacks the extropolability. The deep knowledge repository of temperature profile was yielded based on the determinant mathematical model, which increased the ‘extrapolative ability’ and ‘reliability’. Simulation results have proved that the model presented in this paper is simple, and can satisfy the control requirements.In order to predict and control, A real-time prognosticate model of temperature profile for a reverse flow reactor was built based on dynamic RBFNN with online correcting model parameters. Simulation results have proved that the model is highly accurate and can meet real-time control demand.As an important means that assures normal operations, the cycle period can <WP=8>not only control the bed temperatures effectively, thus adjusting reactive capacity in a wide range, but also be altered easily.In the NIMC strategy proposed, t更多还原