Unknown-length motif discovery methods in environmental monitoring time series

The search for information of interest in massive time series is crucial in many industrial applications. Companies need their data to be analyzed or modeled in real time, which often requires to extract some patterns, also referred as motifs. However, for diverse and ever more signals, human expertise is overwhelmed by time and by huge amount of data. It is the case for environmental monitoring where it is question to detect radiological phenomena from environmental signals. In this paper, we propose an unsupervised and unknown length motif discovery method based on the Matrix Profile with a low computational cost. Its performance is evaluated on a dataset of simulated radiological signals dedicated to environmental monitoring, and compared to a similarity DTW based method and to a classical standard deviation based method. The advantages and drawbacks of each method are highlighted in terms of performance, runtime, accuracy and robustness to different types of noisy signals.