Multilinear domain-specific domain generalization
A multilinear domain-specific domain generalization (MDSDG) approach that utilizes information stored in multilinear indices of data domains to improve machine learning. In particular—based on limited sample size(s) in observed scenarios—an array of models is jointly trained, which advantageously are generalized to a new, unseen scenario, where only domain descriptions in the form of multilinear indices are available.
1 . A multilinear domain-specific domain generalization method (MDSDG) for distributed fiber optic sensing (DFOS), the method comprising:
providing a DFOS system including a DFOS interrogator/analyzer with neural network in optical communication with an optical sensing fiber;
operating the DFOS system and
by the DFOS interrogator/analyzer with neural network:
obtaining training datasets and identifying any conditions in which an individual dataset is obtained;
determining an array of models parameterized by factor model components according to an MDSDG scheme;
assembling a new model according to a chosen tensor decomposition format wherein the new model includes conditions from a target domain in which the new model will be executed wherein at least some of the new model conditions are different from the conditions in which the training datasets were obtained;
obtaining DFOS data in the new model conditions; and
analyzing the DFOS data in the new model conditions using the new model and detecting an environmental condition along the optical sensing fiber.
2 . The method of claim 1 wherein the new model conditions were unavailable during the obtaining of the training datasets.
3 . The method of claim 1 wherein the training datasets include waterfall images resulting from operation of the DFOS system.
4 . A distributed fiber optic sensing system DFOS employing multilinear domain-specific domain generalization (MDSDG) for distributed fiber optic sensing the system comprising:
an interrogator in optical communication with an optical sensing fiber and configured to
interrogate the optical sensing fiber with laser light pulses and receive reflected or scattered light therefrom;
an analyzer with neural network communicatively coupled to the interrogator and configured to:
obtain training datasets from the received reflected or scattered light and identify any conditions in which an individual dataset is obtained;
determine an array of models parameterized by factor model components according to an MDSDG scheme;
assemble a new model according to a chosen tensor decomposition format wherein the new model includes conditions from a target domain in which the new model will be executed wherein at least some of the new model conditions are different from the conditions in which the training datasets were obtained;
obtain DFOS data in new model conditions; and
analyze the DFOS data in the new model conditions using the new model to detect an environmental condition along the optical sensing fiber.
5 . The system of claim 4 wherein the new model conditions are unavailable during the obtaining of the training datasets.
6 . The system of claim 4 wherein the training datasets include waterfall images resulting from operation of the DFOS system.