方程误差类模型的限定记忆加权辨识-控制科学与工程专业论文.docx

Classified Index: TP273 U.D.C: 62 Dissertation for the Master Degree in Engineering WEIGHTED FIXED MEMORY RECURSIVE LEAST SQUARES ALGORITHMS FOR EQUATION ERROR TYPE MODELS Candidate： Yu-Xin Wei Supervisor： Associate Prof. Ai-Guo Wu Academic Degree Applied for： Master of Engineering Speciality： Control Science and Engineering Affiliation： Shenzhen Graduate School Date of Defence： June, 2012 Degree-Conferring-Institution： Harbin Institute of Technology 摘 要 系统辨识是通过系统输入输出数据确定系统数学模型的技术。很多实际系统 的参数是随时间变化的，故对参数时变系统的辨识研究具有重要意义。本文针对 参数时变的方程误差模型、方程误差滑动平均模型、方程误差自回归模型和方程 误差自回归滑动平均模型四类方程误差类模型提出了两种限定记忆的加权辨识方 法。主要研究结果如下： 1) 针对方程误差模型提出了 I 型和 II 型限定记忆加权辨识算法。其中 I 型算法 通过引入中间项来建立相邻两个时刻参数估计的关系，II 型算法直接建立相邻两个 时刻参数估计的关系。 2) 引入常数加权、指数函数加权和线性函数加权三种加权方法。通过比较显 示，指数函数加权与常数加权相比可以更好的保留系统实时信息，指数函数加权 和线性函数加权在一定范围内可以相互替代。 3) 以两种基本的限定记忆加权辨识算法为基础进一步推导出了适用于方程误 差滑动平均模型、方程误差自回归模型、方程误差自回归滑动平均模型的限定记 忆加权辨识算法。针对四种不同类型的时变参数对两种算法进行了数值仿真。仿 真结果显示 II 型算法稍优于 I 型算法，指数函数加权优于遗忘因子加权，指数函 数加权和线性函数加权在一定范围内可以取得相同的辨识结果。 关键词：系统辨识；限定记忆；加权；时变系统 Abstract System identification is a kind of techniques to determine mathematical model by using input and output data. It is important to investigate identification of system with time-varying parameters as the parameters of a real system often change with time. In this thesis, two kinds of weighted fixed memory recursive least squares algorithms for equation error models, equation error moving average models, equation error autogressive models and equation error autogressive moving average models are proposed. The main contributions of the thesis are as follows. Type-I and Type-II weighted fixed memory recursive least squares algorithm for equation error models are proposed. Type-I algorithm is obtained by introducing an intermediate variable to parameter estimation of adjacent time. Type-II algorithm is obtained directly by building up relationship of parameter estimation of adjacent time. Exponential function weighting and linear function weighting are compared to constant weightin