DETECTING AND PROCESSING ANOMALOUS PARAMETER DATA POINTS BY A MOBILE WIRELESS DATA NETWORK FORECASTING SYSTEM

• US 20140111517A1
• Filed: 10/22/2012
• Published: 04/24/2014
• Est. Priority Date: 10/22/2012
• Status: Active Grant

• Current Assignee: United States Cellular Corporation (Telephone & Data Systems, Inc.)
• Original Assignee: United States Cellular Corporation (Telephone & Data Systems, Inc.)
• Inventors: Irizarry, Michael Shannon, Saxena, Narothum, Vela, Mario
• Granted Patent: US 10,531,251 B2
• Time in Patent Office: Days
• US Class Current: 345/440
• CPC Class Codes: H04M 15/80 Rating or billing plans; Tariff determination aspects
• CPC Class Codes: H04M 15/8016 based on quality of service [QoS]
• CPC Class Codes: H04M 15/8027 based on network load situation
• CPC Class Codes: H04W 4/24 Accounting or billing

Abstract

A computer-automated method is described for identifying and remediating anomalous data points for acquired performance parameter (e.g., KPI) values for components of a mobile wireless data network—such as PDSNs. The remediated data point sets is thereafter used as input to a forecasting model rendering a forecast for a mobile wireless network performance indicator/parameter. Data point sequence seasonality is exploited to improve the detection of anomalous data points. Furthermore, correcting data point sets is reiterated until changes between two iterations of the correction process are within a threshold magnitude for terminating the iterative data point sequence correction process.

Claims

1. A method for rendering, by a computerized mobile wireless data network performance parameter forecasting system, a forecast for a mobile wireless network performance parameter, the method comprising the steps of:

acquiring, via a communications network infrastructure for the mobile wireless network, a raw data point set containing a series of data point values for the mobile wireless network parameter;

correcting the raw data point set, the correcting comprising the sub-steps of;

performing a first linear regression on the raw data point set to render an initial best fit line,

calculating a current series of data point residuals based upon;

a current best fit line, the current best fit line being in a first instance the initial best fit line, and

a current series of data point values, the current series of data point values being in a first instance the raw data point set,

decomposing the current series of data point residuals into a current set of seasonal data point residual subsets,

performing a boxplot analysis on the current set of seasonal data point residual subsets to identify a current set of anomalous data point residuals,

correcting the current set of anomalous data point residuals to render a corrected data point residual set, and

applying the corrected data point residual set to the current best fit line to render a current corrected data point set; and

applying the current corrected data point set to a forecast model to render a forecast for the mobile wireless network performance parameter.

2. The method of claim 1 wherein the correcting the raw data point set further comprises:

performing, after the applying step, a further linear regression on the current corrected data point set to render a new best fit line;

comparing the new best fit line to the current best fit line; and

conditionally repeating, based upon the comparing step, the calculating, decomposing, performing the boxplot analysis, correcting the current set of anomalous data point residuals, and applying the corrected data point residual set sub-steps using;

the current corrected data point set as the current series of data point values, and

the new best fit line as the current best fit line.

3. The method of claim 2 wherein the comparing step comprises comparing a slope for the new best fit line and current best fit line.

4. The method of claim 3 wherein the comparing step comprises comparing a difference in magnitude of the respective slope values for the new best fit line and the current best fit line and a threshold magnitude value.

5. The method of claim 3 wherein the comparing step comprises comparing a ratio magnitude for the new best fit line and the current best fit line and a threshold ratio.

6. The method of claim 1 wherein the seasonality of the current set of seasonal data point residual subsets is daily.

7. The method of claim 6 wherein the granularity of the raw data point set is one data point per hour.

8. The method claim 1 wherein the mobile wireless network performance parameter is a measure of throughput on a mobile wireless data network component.

9. The method of claim 8 wherein the mobile wireless data network component is a packet data serving node (PDSN).

10. The method of claim 1 wherein the forecast model renders a forecast value based upon input data point vectors comprising a time-series sequential set of data points.

11. A non-transitory computer-readable medium including computer-executable instructions that, when executed by a computerized mobile wireless data network performance parameter forecasting system, facilitate rendering a forecast for a mobile wireless network performance parameter by performing a method comprising the steps of:

acquiring, via a communications network infrastructure for the mobile wireless network, a raw data point set containing a series of data point values for the mobile wireless network parameter;

correcting the raw data point set, the correcting comprising the sub-steps of;

performing a first linear regression on the raw data point set to render an initial best fit line,

calculating a current series of data point residuals based upon;

a current best fit line, the current best fit line being in a first instance the initial best fit line, and

a current series of data point values, the current series of data point values being in a first instance the raw data point set,

decomposing the current series of data point residuals into a current set of seasonal data point residual subsets,

performing a boxplot analysis on the current set of seasonal data point residual subsets to identify a current set of anomalous data point residuals,

correcting the current set of anomalous data point residuals to render a corrected data point residual set, and

applying the corrected data point residual set to the current best fit line to render a current corrected data point set; and

applying the current corrected data point set to a forecast model to render a forecast for the mobile wireless network performance parameter.

12. The non-transitory computer-readable medium of claim 11 wherein the correcting the raw data point set further comprises:

performing, after the applying step, a further linear regression on the current corrected data point set to render a new best fit line;

comparing the new best fit line to the current best fit line; and

conditionally repeating, based upon the comparing step, the calculating, decomposing, performing the boxplot analysis, correcting the current set of anomalous data point residuals, and applying the corrected data point residual set sub-steps using;

the current corrected data point set as the current series of data point values, and

the new best fit line as the current best fit line.

13. The non-transitory computer-readable medium of claim 12 wherein the comparing step comprises comparing a slope for the new best fit line and current best fit line.

14. The non-transitory computer-readable medium of claim 13 wherein the comparing step comprises comparing a difference in magnitude of the respective slope values for the new best fit line and the current best fit line and a threshold magnitude value.

15. The non-transitory computer-readable medium of claim 13 wherein the comparing step comprises comparing a ratio magnitude for the new best fit line and the current best fit line and a threshold ratio.

16. The non-transitory computer-readable medium of claim 11 wherein the seasonality of the current set of seasonal data point residual subsets is daily.

17. The non-transitory computer-readable medium of claim 16 wherein the granularity of the raw data point set is one data point per hour.

18. The non-transitory computer-readable medium claim 11 wherein the mobile wireless network performance parameter is a measure of throughput on a mobile wireless data network component.

19. The non-transitory computer-readable medium of claim 18 wherein the mobile wireless data network component is a packet data serving node (PDSN).

20. The non-transitory computer-readable medium of claim 11 wherein the forecast model renders a forecast value based upon input data point vectors comprising a time-series sequential set of data points.