時系列データを判別するカオス的手法について

概要

論文の詳細を見る
Several methods that distinguish between a normal and an abnormal time series have been proposed. See Iokibe [3], Kaplan and Glass [4], and Wayland, Bromley, Pickett and Passamante [7]. These methods are algorithmically complicated, and then it is hard to clear the mathematical properties of them. In this paper we propose two simple methods for the problem of classification of time series data, which are called cos analysis method (CAM) and simplified cos analysis method (SCAM). Applying the proposed methods to the artificially produced chaotic time series data and the pressure data of an extruder, we show that we may practically use the methods for checking the strangeness of machines. Furthermore, using ergodic theory, we show that the quantity derived by the simplified cos analysis method equals to -1/2, when the time series data is random.
2002-03-15