一类时变结构系统的辨识及预报

Identification and prediction for a kind of time-varying structure systems

给出一种变阶式拟最小二乘法 ,该算法由主算法和切换算法组成 ,能在线估计模型的参数 ,当系统结构发生变化时 ,能自动改变模型阶数 ,并由切换算法为主算法提供初值 ,从而实现平滑过渡 .在此基础上提出了一种结构适应式自校正预报器 ,该预报器由参数估计器和最优预报器组成 ,能在线、实时、自动改变自身结构和参数 ,并能自动补偿由于模型不精确引起的误差 .用该预报器对某分馏塔温度进行 70步跟踪预报 ,平均相对预报误差为 2 .4 % ,预报精度提高了 0 .16 % .

A class of varying\|order quasi least square algorithm is presented. This algorithm is composed of main and switching algorithms and can be used to estimate the parameters of a model. When the structure of the system varies, the order of the model can also change automatically. The initial values of the main algorithm are given by the switching algorithm. Thus a smooth switching is guaranteed. Meanwhile, a kind of structure\|adaptive self\|tuning predictor is proposed, which consists of parameter estimator and optimal predictor. The structure and parameters of predictor can be changed online real\|time and automatically, and the error produced by inaccurate model can be compensated. This kind of predictor is used to predict the temperature of a fractionating tower by following 70 steps. The average relative error is 2.4%. The accuracy of predicting is improved by 0.16% .