US 12,229,164 B1
Multivariate time-series search and classification learning methods and systems
Brian Parsonnet, Fort Collins, CO (US); Joseph R. DiSantis, Houston, TX (US); and Aswin N. Venkat, Belmont, CA (US)
Assigned to Seeq Corporation, Seattle, WA (US)
Filed by Seeq Corporation, Seattle, WA (US)
Filed on Sep. 16, 2020, as Appl. No. 17/022,772.
Claims priority of provisional application 62/901,075, filed on Sep. 16, 2019.
Int. Cl. G06F 16/28 (2019.01); G06F 16/23 (2019.01); G06F 16/2458 (2019.01); G06N 20/00 (2019.01)
CPC G06F 16/285 (2019.01) [G06F 16/2379 (2019.01); G06F 16/2474 (2019.01); G06N 20/00 (2019.01)] 28 Claims
OG exemplary drawing
 
1. A system comprising:
a processing device; and
a non-transitory computer-readable medium encoded with computer executable instructions that when executed by the processing device cause the processing device to:
given a multi-dimensional ordered data series of sensor data of a monitored system and associated classification data for identified portions of the ordered data series, generate a data classification model by iteratively;
automatically selecting, from a library of detectors each for extracting a type of corresponding characteristic of the ordered data series, a plurality of detectors each extracting a different type of a corresponding characteristic of the ordered data series;
automatically selecting, from a library of classifiers, a plurality of classifiers each generating a prediction outcome for identifying a portion of the ordered data series corresponding to the associated classification data, wherein each of the plurality of classifiers is provided the extracted types of corresponding characteristics of the ordered data series determined from the selected plurality of detectors;
determining a relative accuracy of the prediction outcomes from each of the plurality of classifiers to the portion of the ordered data series; and
altering the data classification model based on the determined relative accuracy of the prediction outcomes from each of the plurality of classifiers to the portion of the ordered data series, wherein altering the data classification model comprises selecting a second plurality of detectors from the library of detectors and a second plurality of classifiers from the library of classifiers;
apply a data set of streamed sensor data to the data classification model to identify a portion of the data set of streamed sensor data corresponding to the identified portions of the ordered data series; and
conduct, based on the identified portion of the data set of streamed sensor data, a remedial action of the monitored system.