Abstract: This paper proposed a new prediction method for outliers over data stream based on sparse representation to improve the optimum prediction speed and performance of outliers over data stream. Combining the wavelet noise detection method, using newly developed tools for sparse representation, a transformation method for outliers over data stream was proposed. In order to identify outliers, the introduction of random measurement matrix of wavelet transform coefficients was applied with sparse representation to forecast data value in the future timestamp. The simulation results on actual data source show that this method can provide precise instantaneous detection under certain conditions.

Key words: data streams, outlers prediction, sparse representation, wavelet transform

摘要： 为了提高数据流中异常数据的预测速度与精度,提出一种基于稀疏表示的数据流异常数据预测方法。结合了小波噪声检测方法,采用新近发展起来的稀疏表示工具,对含有异常数据的数据流进行小波变换,并得到一组小波系数,然后对这些系数建立稀疏表示模型。引入随机测量矩阵对小波系数进行变换,恢复小波系数的稀疏性达到预测异常数据的目的。仿真结果表明,在一定条件下该方法可获得相当好的预测效果。

关键词: 数据流, 异常数据预测, 稀疏表示, 小波变换