DEVICE CHARACTERIZATION SYSTEM AND METHODS

• US 20130027220A1
• Filed: 07/29/2011
• Published: 01/31/2013
• Est. Priority Date: 07/29/2011
• Status: Active Grant

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Marwah, Manish, Patel, Chandrakant, Lyon, Geoff M., Lyons, Martha L., Arlitt, Martin, Bash, Cullen E.
• Granted Patent: US 8,836,536 B2
• Time in Patent Office: Days
• US Class Current: 340/870.16
• CPC Class Codes: H04Q 2209/845 where the measuring is synchronized between sensing devices
• CPC Class Codes: H04Q 9/00 Arrangements in telecontrol or telemetry systems for selectively calling a substation from a main station, in which substation desired apparatus is selected for applying a control signal thereto or for obtaining measured values therefrom

Abstract

In an embodiment, a device characterization system includes a sensor to sense an attribute of a device, a processor, and an algorithm executable on the processor to collect time series data of the attribute from the sensor, detect edges in the data, identify clusters from the edges, label the clusters based on input from a supervisor, and estimate device characterization parameters from the clusters.

Claims

1. A method of characterizing temporal behavior of a device, comprising:

collecting time series data for an attribute of the device;

pre-processing the time series data;

detecting edges in the time series data;

arranging groups of similar edges into clusters;

receiving label input from a supervisor;

labeling each cluster as being associated with a physical event of the device based on the label input;

based on the clusters, estimating characterization parameters that characterize normal operation of the device.

2. A method as in claim 1, further comprising:

monitoring ongoing time series data of the attribute;

comparing the characterization parameters with operational parameters from the ongoing time series data;

detecting abnormal operation of the device based on the comparing;

providing a labeled alert indicating a physical event related to the abnormal operation of the device.

3. A method as in claim 1, wherein collecting time series data comprises sensing power consumption of the device at a specific sampling rate.

4. A method as in claim 1, wherein pre-processing the time series data comprises removing invalid values, filling in missing values, making time intervals uniform between data samples, filtering the data to remove noise and outlying data values.

5. A method as in claim 1, wherein detecting edges comprises detecting rising and falling edges of the time series data.

6. A method as in claim 1, wherein detecting edges provides a time series of all deltas in successive time points of the attribute.

7. A method as in claim 1, wherein arranging groups of similar edges into clusters comprises determining a number of clusters.

8. A method as in claim 1, wherein arranging groups of similar edges into clusters comprises clustering the groups with a clustering algorithm.

9. A method as in claim 8, wherein the clustering algorithm comprises a K-means algorithm.

10. A method as in claim 1, wherein estimating characterization parameters comprises computing the parameters through a density estimation using a mixture model.

11. A method of characterizing temporal behavior of a device, comprising:

collecting time series data for an attribute of the device;

detecting edges in the time series data;

arranging groups of similar edges into a number of clusters;

receiving a different number of clusters from a supervisor;

arranging the groups of similar edges into the different number of clusters;

labeling each cluster with a label received from the supervisor; and

estimating characterization parameters from the clusters that characterize normal operation of the device.

12. A device characterization system comprising:

a sensor to sense an attribute of a device;

a processor;

a semi-supervised, characterization algorithm executable on the processor to collect time series data of the attribute from the sensor, detect edges in the data, identify clusters from the edges, label the clusters based on input from a supervisor, and estimate device characterization parameters from the clusters.

13. A device characterization system as in claim 12, further comprising:

a server that is remote from the device;

wherein the device comprises the sensor, and the server comprises the processor and the characterization algorithm.

14. A device characterization system as in claim 12, wherein the device comprises the sensor, the processor, and the characterization algorithm.

15. A device characterization system as in claim 12, wherein the characterization algorithm comprises:

a data collection module to collect the time series data;

a data pre-processing module to eliminate noise and invalid values in the data, and to fill in missing values in the data;

an edge detection module to detect the edges in the data;

a clustering module to identify groups of similar edges and arrange the groups of similar edges into the clusters;

a cluster labeling module to receive the input from the supervisor and to label each cluster as being associated with a physical event of the device based on the input; and

a parameter estimation module to estimate the device characterization parameters.

16. A device characterization system as in claim 15, wherein the characterization algorithm further comprises an anomaly detection module to detect abnormal device behavior by comparing the device characterization parameters with operational parameters from ongoing time series data, and to provide an alert with a label indicating a physical event related to the abnormal device behavior.