Distributed acoustic sensing (DAS) system for acoustic event detection using machine learning network selected by game theoretic model and related methods
A distributed acoustic sensing (DAS) system may include an optical fiber, a phase-sensitive OTDR (ϕ-OTDR) coupled to the optical fiber, and a processor cooperating with the ϕ-OTDR. The processor may be configured to train a plurality of machine learning networks with DAS data from the ϕ-OTDR based upon different respective optimizers, select a trained machine learning network from among the plurality thereof based upon a game theoretic model, and generate an acoustic event report from the DAS data using the selected trained machine learning network.
1 . A distributed acoustic sensing (DAS) system comprising:
an optical fiber;
a phase-sensitive optical time domain reflectometer (φ-OTDR) coupled to the optical fiber; and
a processor cooperating with the φ-OTDR and configured to
train a plurality of machine learning networks with DAS data from the φ-OTDR based upon different respective optimizers,
select a trained machine learning network from among the plurality of trained machine learning networks based upon a game theoretic model, and
generate an acoustic event report from the DAS data using the selected trained machine learning network.
2 . The DAS system of claim 1 wherein the plurality of machine learning networks comprises a plurality of Long Short Term Memory (LSTM) networks.
3 . The DAS system of claim 1 wherein the plurality of machine learning networks comprises a plurality of U-Net convolutional neural networks (CNNs).
4 . The DAS system of claim 1 wherein the plurality of machine learning networks comprises a plurality of You Only Look Once (YOLO) networks.
5 . The DAS system of claim 1 wherein the different optimizers comprise Adaptive Moment Estimation (ADAM), Stochastic Gradient Descent with Momentum (SGDM), and Root Mean Square Propagation (RMSProp) deep learning models.
6 . The DAS system of claim 1 wherein the processor is further configured to generate a series of covariance matrices for the DAS data, and select a subset of the DAS data for training the plurality of machine learning networks based upon comparing the series of covariance matrices with a corresponding Toeplitz matrix.
7 . The DAS system of claim 1 wherein the processor is further configured to localize subsets of channels in time for corresponding acoustic events.
8 . The DAS system of claim 1 wherein the processor is further configured to classify different regions within the DAS data using different respective acoustic event classes.
9 . A distributed acoustic sensing (DAS) device comprising:
a phase-sensitive optical time domain reflectometer (φ-OTDR) to be coupled to an optical fiber; and
a processor cooperating with the φ-OTDR and configured to
train a plurality of machine learning networks with DAS data from the φ-OTDR based upon different respective optimizers,
select a trained machine learning network from among the plurality of trained machine learning networks based upon a game theoretic model, and
generate an acoustic event report from the DAS data using the selected trained machine learning network.
10 . The DAS device of claim 9 wherein the plurality of machine learning networks comprises a plurality of Long Short Term Memory (LSTM) networks.
11 . The DAS device of claim 9 wherein the plurality of machine learning networks comprises a plurality of U-Net convolutional neural networks (CNNs).
12 . The DAS device of claim 9 wherein the plurality of machine learning networks comprises a plurality of You Only Look Once (YOLO) networks.
13 . The DAS device of claim 9 wherein the different optimizers comprise Adaptive Moment Estimation (ADAM), Stochastic Gradient Descent with Momentum (SGDM), and Root Mean Square Propagation (RMSProp) deep learning models.
14 . The DAS device of claim 9 wherein the processor is further configured to generate a series of covariance matrices for the DAS data, and select a subset of the DAS data for training the plurality of machine learning networks based upon comparing the series of covariance matrices with a corresponding Toeplitz matrix.
15 . The DAS device of claim 9 wherein the processor is further configured to localize subsets of channels in time for corresponding acoustic events, and classify different regions within the DAS data using different respective acoustic event classes.
16 . A distributed acoustic sensing (DAS) method comprising:
training a plurality of machine learning networks using a processor with DAS data from a phase-sensitive optical time domain reflectometer (φ-OTDR) coupled to an optical fiber based upon different respective optimizers;
selecting a trained machine learning network from among the plurality of trained machine learning networks using the processor based upon a game theoretic model; and
generating an acoustic event report from the DAS data using the processor and the selected trained machine learning network.
17 . The method of claim 16 wherein the plurality of machine learning networks comprises a plurality of Long Short Term Memory (LSTM) networks.
18 . The method of claim 16 wherein the plurality of machine learning networks comprises a plurality of U-Net convolutional neural networks (CNNs).
19 . The method of claim 16 wherein the plurality of machine learning networks comprises a plurality of You Only Look Once (YOLO) networks.
20 . The method of claim 16 wherein the different optimizers comprise Adaptive Moment Estimation (ADAM), Stochastic Gradient Descent with Momentum (SGDM), and Root Mean Square Propagation (RMSProp) deep learning models.
21 . The method of claim 16 further comprising generating a series of covariance matrices for the DAS data using the processor, and selecting a subset of the DAS data for training the plurality of machine learning networks using the processor based upon comparing the series of covariance matrices with a corresponding Toeplitz matrix.
22 . The method of claim 16 further comprising localizing subsets of channels in time for corresponding acoustic events using the processor, and classifying different regions within the DAS data using different respective acoustic event classes using the processor.