Abstract
Distributed acoustic sensing technology is increasingly being used to support production and well management within the oil and gas sector, for example to improve flow monitoring and production profiling. This sensing technology is capable of recording substantial data volumes at multiple depths within an oil well, giving unprecedented insights into production behaviour. However the technology is also prone to recording periods of anomalous behaviour, where the same physical features are concurrently observed at multiple depths. Such features are called ‘stripes’ and are undesirable, detrimentally affecting well performance modelling. This paper focuses on the important challenge of developing a principled approach to identifying such anomalous periods within distributed acoustic signals. We extend recent work on classifying locally stationary wavelet time series to an online setting and, in so doing, introduce a computationally-efficient online procedure capable of accurately identifying anomalous regions within multivariate time series.
Highlights
The ability to accurately analyse geoscience data at, or close to, real time is becoming increasingly important
Due to the high frequency nature of acoustic sensing data, we focus our attention on wavelet-based methods such as the Locally Stationary Wavelet (LSW) processes, introduced by Nason et al (2000)
We focus on the recently proposed multivariate locally stationary wavelet (MvLSW) framework introduced by Park et al (2014), which we later use to model the Distributed Acoustic Sensing (DAS) data described in Sect
Summary
The ability to accurately analyse geoscience data at, or close to, real time is becoming increasingly important. Wavelet approaches to modelling time series have become very popular in recent years, principally because of their ability to provide time-localised measures of the spectral content inherent within many contemporary data (e.g. Killick et al 2013; Nam et al 2015; Chau and von Sachs 2016; Nason et al 2017) This locally stationary modelling paradigm is flexible enough to represent a wide range of non-stationary behaviour and has been extended to enable the modelling and estimation of multivariate non-stationary time series structures (e.g. Sanderson et al 2010; Park et al 2014). The novel contribution in this article is to employ the MvLSW modelling framework of Park et al (2014) to represent the DAS data, using a moving window approach, thereby extending previous work to the online dynamic classification setting This modelling framework allows us to classify multivariate time series with complex dependencies both within and between channels of the series, including those which exhibit visually subtle changes in behaviour over time.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
