输入数据缺失情况下的OE模型辨识算法研究

针对辨识技术应用过程中出现的一类输入数据随机缺失时的辨识问题进行了研究。针对输出误差(OE)模型描述的一类系统,提出了一种模型辨识和缺失数据预测交互迭代的辨识算法。在模型辨识中采用了递推的辨识算法便于形成实时更新的在线辨识策略;而在缺失数据的预测过程中,利用小波降噪技术对预测数据进行适应性的滤波。仿真和分析表明,所提出的辨识算法对连续性输入数据缺失具有很强的鲁棒性;与没有小波技术参与的辨识算法相比,该算法具有较高的模型辨识精度和对缺失数据较好的预测能力。

The research on an algorithm for identification of OE models with missing input data

A research is carried out to deal with the identification using randomly missing input data, which is probably encountered during the application of an identification technique. Aiming at the true system which can be characterized as an output error (OE) model, an interactively iterative identification algorithm consisting of model identification and prediction for missing data, is proposed. During the model identification, a recursive identification algorithm is applied to achieving a real-time and on-line update. In the course of the prediction for missing data, a wavelet de-noising technique is employed to adaptively filter the predicted missing data. Finally, a numerical simulation shows that the proposed algorithm has strong robustness for a segment of missing input data. Compared with the related algorithm without wavelet, the proposed one is capable of giving higher model accuracy and has better prediction ability for missing data.