基于和声搜索的自适应滤波算法

针对子空间辨识方法,从状态空间模型向I/O差分方程转化的取向,使有关矩阵的维数增加,计算复杂度增大的问题,提出了用和声搜索算法辨识CARMA(Controlled Auto-Regressive Moving Average)模型子阶、时滞和参数的方案.该方案基于所提出的输入输出模型转化为状态空间模型的定理,使相应的状态空间模型在CARMA模型辨识之后也被辨识.以油井热洗为例,进行了自适应滤波估计.估计结果表明:该算法有较高的滤波和预报的精度,滤波和一步预报的误差方差的最大值在0.01以内,与子空间辨识方案相比,本方案的计算量约为前者的五分之一.

Adaptive identification and filtering based on harmony search

In subspace identification methods, the means from transforming state space model to I/O difference equation increase the dimension of involved matrixes and calculational burden. To solve this problem, a scheme that estimates the suborders, time-lags and parameters of CARMA(Controlled Auto-Regressive Moving Average) model was proposed by the harmony search algorithm. Based on the presented theorem of transforming I/O model to state space model, corresponding state space model has been identified by this scheme after CARMA model is identified. The oil well heat wash was taken as example to do adaptive filtering estimation. The result shows that this algorithm has very high precision of filtering and prediction, and their maximums of error square difference are within 0.01. Compared with subspace identification methods, the calculational burden of this scheme is 1/5 of the former.