Hierarchic Adaptive Genetic Algorithm Detection Of Outliers In Multivariate Time Series Based On Projection Pursuit

The detection of time series outliers has a very important application in quality con-trol,signal processing,climate change and other practical problems.It is an important issue in the study of time series statistical inference and has been paid much attention by many scholars.However,most of the current mainstream detection methods are aimed at univari-ate time series,and there are few studies on the detection of outliers in multivariate time series.This paper presents a hierarchical adaptive genetic algorithm for detecting outliers in multivariate time series based on projection pursuit.By maximizing the kurtosis coefficient of projected sequences,we found a series of orthogonal optimal projection directions,in which the abnormal effects of outliers were significantly amplified to facilitate the detec-tion of outliers,and the detection of outliers in the original sequence was also converted to the detection of some univariate projection sequences.On the projection sequence,we detect outliers by genetic algorithm.Different from previous detection methods,genetic algorithm,as an optimization search algorithm,can compare a large number of outlier pat-terns at the same time,so as to detect the location and type of outliers(AO and IO)at the same time.Aiming at the disadvantages of the basic genetic algorithm,such as low search-ing efficiency and growing time of the adaptive genetic algorithm,a hierarchical adaptive genetic algorithm is adopted to improve the algorithm and the elitist strategy is combined to ensure the retention of the best individual in each generation.Finally,three groups of simulated data and two groups of real data verify that the method presented in this paper is more advantageous than the basic genetic algorithm in search efficiency and computation time in detecting outliers in multivariate time series.