Pattern discovery in time series using autoencoder in comparison to nonlearning approaches

Abstract

In technical systems the analysis of similar situations is a promising technique to gain information about the system’s state, its health or wearing. Very often, situations cannot be defined but need to be discovered as recurrent patterns within time series data of the system under consideration. This paper addresses the assessment of different approaches to discover frequent variable-length patterns in time series. Because of the success of artificial neural networks (NN) in various research fields, a special issue of this work is the applicability of NNs to the problem of pattern discovery in time series. Therefore we applied and adapted a Convolutional Autoencoder and compared it to classical nonlearning approaches based on Dynamic Time Warping, based on time series discretization as well as based on the Matrix Profile. These nonlearning approaches have also been adapted, to fulfill our requirements like the discovery of potentially time scaled patterns from noisy time series. We showed the performance (quality, computing time, effort of parametrization) of those approaches in an extensive test with synthetic data sets. Additionally the transferability to other data sets is tested by using real life vehicle data. We demonstrated the ability of Convolutional Autoencoders to discover patterns in an unsupervised way. Furthermore the tests showed, that the Autoencoder is able to discover patterns with a similar quality like classical nonlearning approaches.

Unknown-length motif discovery methods in environmental monitoring time series

“…Crossmatch [10], [14] is based on the computation of a similarity matrix v. We here use the same formula than Noering et al [4] to get the similarity matrix v. Each value of v(i, j) depends on the previous values of the matrix v:…”

“…Among recent methods of the state-of-art [4], approaches based on discretization [5], autoencoder [6], Dynamic Time Warping (DTW) [7] and Matrix Profile (MP) [8] have been studied. Following on from results presented in the comparative study in [4], approaches based on discretization and autoencoder have not been selected here due to their low performance compared to DTW [7] based approach. Moreover, the MP has shown its flexibility to many problems in data mining [8], [9], which makes it interesting for our problem.…”

“…The latter computes a similarity matrix equivalent to the DTW distance, for two evolving time series. Noering et al [4] adapted the similarity matrix of the Crossmatch algorithm and combined it with the motif extraction and clustering method of [11], to discover motifs in one signal. Our modification consists in thresholding the diagonal of the similarity matrix.…”

Recognizing human activities in Industry 4.0 scenarios through an analysis-modeling- recognition algorithm and context labels

“…-Finally, other artificial intelligence models have been developed, especially for camera-based systems and computer vision. Gaussian models [36], semantic technologies [18], intelligent encoders [5], optimization functions [19] or estimation techniques [20] have been reported very recently. All these approaches have the advantage of showing a very good performance and precision, but they are not flexible Table 1 presents and analyzes works on these scenarios.…”

“…The second EMA preserves the trends in signals Eq. (5). And the third and final EMA must preserve the seasonal information Eq.…”