Sequential detection of common transient signals in high dimensional data stream
研究了在高维数据流中检测瞬态信号的方法，给出了EWMA和MA控制图在单序列中检测瞬态信号的性能近似，并提出了多元EWMA和Min-δ程序以提高检测能力，用道琼斯30工业股票价格数据进行了说明。
Abstract Motivated from sequential detection of transient signals in high dimensional data stream, we first study the performance of EWMA and MA charts for detecting a transient signal in a single sequence in terms of the power of detection under the constraint of false detecting probability in the stationary state. Satisfactory approximations are given for the false detection probability and the power of detection. Comparison of EWMA, MA, and CUSUM charts shows that both charts are quite competitive. A multivariate EWMA procedure is considered by using the squared sum of individual EWMA processes and a fairly accurate approximation for the false detection probability is also given. To increase the power of detection, we use the Min‐δ procedure by truncating the estimated weak signals. Dow Jones 30 industrial stock prices are used for illustration.