【作者】 黄宇；

【导师】 刘长良； 董泽；

【作者基本信息】 华北电力大学 ， 控制理论与控制工程， 2012， 博士

【摘要】 人们在认识和改造客观世界的过程中,总存在一些难以用现有的知识去定量描述的系统,系统辨识就是根据测量系统产生的各种信号去构造系统的模型,它是联系现实和数学模型的纽带。在优化计算方面,量子计算相比于经典优化计算,在某些方面可能拥有后者无法比拟的优势。本文利用量子计算与量子优化的方法对热工过程进行了辨识,分别从线性单入单出系统传递函数模型的辨识、多变量子空间模型辨识和非线性神经网络模型辨识等三个方面进行了研究。主要创新成果有：(1)针对量子粒子群算法(QPSO)的收敛速度和寻优精度问题,提出了一种改进的QPSO算法。首先,采用混沌序列初始化粒子的初始角位置；其次,在算法中加入变异处理,有效地增加了种群的多样性,避免早熟收敛。函数优化测试结果表明：本文提出的算法具有良好的优化效果。同时利用本文提出的算法对经典的具典型意义的传递函数族模型进行了辨识,辨识结果证明了这种算法的有效性。利用此算法,在结合某DCS的基础上,编制出了一种通用的热工对象模型辨识算法模块,并应用于某循环流化床电厂的辨识,取得了令人满意的辨识结果。(2)用实例证明了状态子空间辨识方法是一种有差辨识方法。为了获得辨识参数的一致无偏估计,在经典状态子空间辨识的基础上,提出了基于优化算法的两段辨识方法.。首先利用经典状态子空间辨识获取被辨识对象的初始信息,然后利用改进的量子粒子群算法对其进行优化,通过实例验证了本文所提出算法的有效性。最后对某电厂协调控制系统的数据进行辨识。辨识的结果表明：本文所提出的方法可以适用于工业过程多变量系统的辨识,且具有良好的辨识精度。(3)量子遗传算法是基于量子计算原理的概率优化方法,在量子门更新过程中,旋转角的大小直接影响优化的结果和进化的速度。本文针对模糊量子遗传算法(FQGA)容易导致系统陷入局部最优的缺点,将量子衍生交叉算法的思想引入FQGA,提出了一种新的量子遗传算法。同时利用该方法构造径向基函数神经网络并进行非线性系统的辨识。其特点是通过这种新的量子遗传算法实现对RBF神经网络权值、宽度和中心位置等有关参数的估计,其速度快、精度高,从而通过RBF神经网络有效地完成了对非线性系统的辨识。对典型非线性函数辨识的测试表明：该方法有效的提高了量子遗传算法的计算精度和收敛速度。同时利用这种方法设计了一种通用的热上对象模型辨识神经网络算法,并编制了专用的模型识别软件,对某电厂循环流化床锅炉一次风对床温的动态特性进行辨识,结果表明该方法是一种精度比较高的辨算法,具有一定的实用价值。更多还原

【Abstract】 There are always some sytems that people cannot describe them by using existing knowledge during the process of recognizing and transforming the objective world. System identification is the link between reality and mathematical model, which can be defined as the way of systematic modeling according to measured serious systematic signals. In the term of optimization calculation, quantum computing has incomparable advantages over classic optimization calculation in some aspects.This article identifies thermal process by using the way of quantum computing and quantum optimization, and studies linear and single-output system transfer function model identification, multivariable subspace identification and nonlinear neural network identification and so on. The main innovative efforts are:(1) In order to improve convergence speed and precision of optimization in quantum particle swarm optimization (QPSO), an improved QPSO algorithm was presented. First, chaotic sequences are used to initialize the origin angle position of particle; Second, mutation algorithm is introduced, which can effectively increase diversity of population, and also can avoid premature convergence. The test results of function optimization show that the proposed algorithm has better optimized effect. The improved algorithm proposed in this paper was applied to identify the classic adaptive IIR model, and results proved the validity of the algorithm. On the basis of DCS, a general-purpose identification algorithm modular for thermal object model is programmed, and it is applied to the identification of circulating fluidized bed power plant, achieving satisfactory results.(2)We use examples to prove that the state subspace identification method is a kind of discriminating identification method. In order to obtain consistent unbiased estimation parameters, two sections identification method is puts forward on the basis of classical state subspace identification and optimization algorithm. First, the initial values of the object are identified by using classical state subspace identification, then we use improved quantum partial swarm algorithm to get the consistent unbiased estimation parameters. Examples show the effectiveness of the presented algorithm. At last, a coordinated control system in power plant is identified, and results showed that the presented method can be used in identification MIMO system in industrial process, and it can get good results.(3) Quantum genetic algorithm is a probability optimization method which is based on quantum compute principle. The precision and the rate of convergence are impacted by rotation angle. Aiming at the shortcoming of fuzzy quantum genetic algorithm (FQGA), quantum-inspired crossover method was introduced to FQGA, and a novel quantum genetic algorithm was put forward. Using this method, an identification algorithm of nonlinear systems is presented. This method is characterized by estimating parameters such as weight, width and central position of RBF NN using the new quantum genetic algorithm. High velocity and accuracy of the method enable nonlinear systems to be efficiently identified by using RBF NN. The results of identifying typical nonlinear function demonstrate that the precision and the rate of convergence are improved. A special program was compiled to identify the object model of the thermal process, and the dynamic process between primary air feed rate and bed temperature was identified. The results show that accuracy of the approach is high and has a certain practical value.更多还原