Detection of anomalous utility usage

• US 9,535,917 B1
• Filed: 09/28/2012
• Issued: 01/03/2017
• Est. Priority Date: 09/28/2012
• Status: Active Grant

• Current Assignee: Emc IP Holding Company LLC (Dell Technologies Inc.)
• Original Assignee: EMC Corporation (Dell Technologies Inc.)
• Inventors: Lin, Derek
• Primary Examiner(s): Vital, Pierre
• Assistant Examiner(s): Sultana, Nargis
• Application Number:
  US13/631,298
• Time in Patent Office: 1,558 Days
• Field of Search: 707/758, 707/688, 707/690, 707/692, 707/700
• US Class Current: 1/1
• CPC Class Codes: G06F 16/148 File search processing
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06Q 50/06 Energy or water supply

Abstract

Detecting for anomalous utility usage, including: determining with respect to the subject set of utility usage data a portion that is not associated with a predetermined set of significant components; determining that the portion that is not associated with the predetermined set of significant components exceeds a prescribed threshold; and concluding, based at least in part on the determination that the portion that is not associated with the predetermined set of significant components exceeds the prescribed threshold, that the subject set of utility usage data is anomalous.

Claims

1. A system of anomalous utility usage detection, comprising:

a memory or other storage device configured to store a subject set of utility usage data; and

a processor coupled to the memory or other storage device and configured to;

analyze a set of historical utility usage data associated with one or more readings over a particular window of time and derive a plurality of components from the set of historical utility usage data, wherein each of the plurality of components represents a characteristic of the set of historical utility usage data;

select a subset of the plurality of components to use as a set of significant components, wherein the plurality of components is associated with a corresponding plurality of significance values, wherein the subset of the plurality of components is selected based at least in part on the corresponding plurality of significance values, wherein the set of significant components is selected to represent normal utility usage data;

determine with respect to the subject set of utility usage data a portion that is not associated with the set of significant components;

determine that the portion that is not associated with the set of significant components exceeds a prescribed threshold, wherein the portion that is not associated with the set of significant components corresponds to a difference between the subject set of utility usage data and a projection of the subject set of utility usage data onto a subspace spanned by the set of significant components; and

conclude, based at least in part on the determination that the portion that is not associated with the set of significant components exceeds the prescribed threshold, that the subject set of utility usage data is anomalous; and

wherein, the plurality of components is decomposed from the set of historical utility usage data using a principal component analysis that comprises;

determining one or more eigenvectors associated with a training matrix generated using the set of historical usage data;

determining one or more eigenvalues respectively corresponding to the one or more eigenvectors; and

selecting the set of significant components based at least in part on the one or more eigenvalues respectively corresponding to the one or more eigenvectors.

2. The system of claim 1, wherein the subject set of utility usage data is associated with a meter and the set of significant components is associated with the meter.

3. The system of claim 1, wherein the subject set of utility usage data is associated with a group of meters and the set of significant components is associated with the group of meters.

4. The system of claim 1, wherein the corresponding plurality of significance values associated with the plurality of components comprises a plurality of variances.

5. The system of claim 4, wherein the processor is further configured to receive a configuration associated with a desired percentage of variance to use to select the subset of the plurality of components.

6. The system of claim 1, wherein the set of historical utility usage data comprises a series of recorded meter readings and corresponding time indications.

7. The system of claim 1, wherein the set of significant components is determined based at least in part on a transformation applied to the set of historical utility usage data.

8. The system of claim 1, wherein the prescribed threshold is determined based at least in part on a portion of the set of historical utility usage that is outside of the set of significant components.

9. The system of claim 1, wherein the processor is further configured to display an alert associated with information associated with the determination of the subject set of utility usage data as being anomalous.

10. A method of anomalous utility usage detection, comprising:

analyzing a set of historical utility usage data associated with one or more readings over a particular window of time and deriving a plurality of components from the set of historical utility usage data, wherein each of the plurality of components represents a characteristic of the set of historical utility usage data;

selecting a subset of the plurality of components to use as a set of significant components, wherein the plurality of components is associated with a corresponding plurality of significance values, wherein the subset of the plurality of components is selected based at least in part on the corresponding plurality of significance values, wherein the set of significant components is selected to represent normal utility usage data;

determining, using one or more processors, with respect to the subject set of utility usage data a portion that is not associated with the set of significant components;

determining that the portion that is not associated with the set of significant components exceeds a prescribed threshold, wherein the portion that is not associated with the set of significant components corresponds to a difference between the subject set of utility usage data and a projection of the subject set of utility usage data onto a subspace spanned by the set of significant components; and

concluding, based at least in part on the determination that the portion that is not associated with the set of significant components exceeds the prescribed threshold, that the subject set of utility usage data is anomalous; and

wherein, the plurality of components is decomposed from the set of historical utility usage data using a principal component analysis that comprises;

determining one or more eigenvectors associated with a training matrix generated using the set of historical usage data;

determining one or more eigenvalues respectively corresponding to the one or more eigenvectors; and

selecting the set of significant components based at least in part on the one or more eigenvalues respectively corresponding to the one or more eigenvectors.

11. The method of claim 10, wherein the subject set of utility usage data is associated with a meter and the set of significant components is associated with the meter.

12. The method of claim 10, wherein the subject set of utility usage data is associated with a group of meters and the set of significant components is associated with the group of meters.

13. The method of claim 10, wherein the corresponding plurality of significance values is associated with the plurality of components comprises a plurality of variances.

14. The method of claim 13, further comprising receiving a configuration associated with a desired percentage of variance to use to select the subset of the plurality of components.

15. The method of claim 10, wherein the set of historical utility usage data comprises a series of recorded meter readings and corresponding time indications.

16. The method of claim 10, wherein determining the portion that is not associated with the set of significant components includes determining a degree to which the subject set of utility usage data deviates from the set of significant components.

17. The method of claim 10, wherein the training matrix is generated using the set of historical utility usage data.

18. The method of claim 10, further comprising:

in response to concluding that the subject set of utility usage data is anomalous, sending, to a user, an alert indicating that the subject set of utility usage data is anomalous.

19. A computer program product for anomalous utility usage detection, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:

analyzing a set of historical utility usage data associated with one or more readings over a particular window of time and deriving a plurality of components from the set of historical utility usage data, wherein each of the plurality of components represents a characteristic of the set of historical utility usage data;

selecting a subset of the plurality of components to use as a set of significant components, wherein the plurality of components is associated with a corresponding plurality of significance values, wherein the subset of the plurality of components is selected based at least in part on the corresponding plurality of significance values, wherein the set of significant components is selected to represent normal utility usage data;

determining with respect to the subject set of utility usage data a portion that is not associated with the set of significant components;

determining that the portion that is not associated with the set of significant components exceeds a prescribed threshold, wherein the portion that is not associated with the set of significant components corresponds to a difference between the subject set of utility usage data and a projection of the subject set of utility usage data onto a subspace spanned by the set of significant components; and

concluding, based at least in part on the determination that the portion that is not associated with the set of significant components exceeds the prescribed threshold, that the subject set of utility usage data is anomalous; and

wherein, the plurality of components is decomposed from the set of historical utility usage data using a principal component analysis that comprises;

determining one or more eigenvectors associated with a training matrix generated using the set of historical usage data;

determining one or more eigenvalues respectively corresponding to the one or more eigenvectors; and

selecting the set of significant components based at least in part on the one or more eigenvalues respectively corresponding to the one or more eigenvectors.

20. The computer product of claim 19, further comprising instructions for:

generating the training matrix using the set of historical utility usage data.