Processing

Please wait...

Settings

Settings

Goto Application

1. WO2007089285 - MODELING OF TRANSACTION FLOW FOR FAULT DETECTION

Note: Text based on automatic Optical Character Recognition processes. Please use the PDF version for legal matters

[ EN ]

What is claimed is:

1. A method for detecting faults in a distributed transaction system comprising:
receiving data corresponding to flow intensity measured at a plurality of monitoring points in the distributed transaction system during normal operation of the system;
generating a model of flow dynamics in the distributed transaction system by automatically deriving a relationship that characterizes a normal flow intensity through a segment of the distributed transaction system; and
monitoring the distributed transaction system by detecting deviations from said model of flow dynamics.

2. The method according to claim 1 wherein said model of flow dynamics is validated by inputting new flow intensity data for said segment and performing sequential testing to said derived relationship to derive a fitness score and wherein said fitness score is used to evaluate the credibility of said derived relationship as a confidence score.

3. The method according to claim 2 wherein said deviations from said model of flow dynamics are detected by deriving a residual by tracking conformance between observed flow intensity measurements for said segment and an output of said model for that segment.

4. The method according to claim 3 further comprising organizing said received flow intensity data to characterize said segment between two monitoring points wherein said segment comprises at least one component.

5. The method according to claim 4 further comprising determining if said residual is abnormal if it is above a threshold.

6. The method according to claim 5 further comprising correlating said confidence score with said residual to evaluate credibility of said residual when used to detect transaction system faults.

7. The method according to claim 6 further comprising correlating said residual with its components to isolate transaction system faults.

8. The method according to claim 7 wherein said model is a regression model.

9. The method according to claim 8 wherein said regression model further comprises:
initializing a prediction model that predicts a flow intensity output for a segment based on past flow intensity inputs and outputs for said segment;
inputting a sequence of orders for said prediction model;
for each said order, over a first predetermined period of time;
deriving an observation using said prediction model wherein flow intensity data for said segment is input;
for each said observation, calculating a simulated output based on said observation flow intensity output;
defining an estimation error for each said observation output with an actual flow intensity output for said segment;
deriving a model parameter that minimizes said estimation error by quantitatively estimating the trend of said observation outcomes; and for each said model parameter, deriving a fitness score over said first predetermined period of time.

10. The method according to claim 9 further comprising:
testing each said model parameter for said first predetermined period of time wherein flow intensity data is input to validate each said model parameter as a model based on whether each said fitness score is above a predetermined fitness threshold;
calculating a confidence score for each said model parameter by counting the number of times each said model fitness score is higher than said fitness threshold over a predetermined series of time windows;
using said model for fault detection so long as said confidence score for said model is above a predetermined confidence threshold; and
updating each said model confidence score over time.

11. The method according to claim 10 ftirther comprising:
deriving a residual for said model, wherein said residual is the difference between an observed segment input and output and a predicted output from said model;
deriving a residual threshold based upon past values of said residual; and
comparing said residual against said residual threshold to detect said deviations.

12. The method according to claim 7 wherein said model is a Gaussian distribution model.

13. The method according to claim 12 wherein said Gaussian model further comprises:
initializing a probability density function to approximate a real data distribution;
correlating pairs of flow intensity data measurements as a time series;

deriving a two-dimensional Gaussian distribution;
deriving a mixture parameter set;
tracking the mass characteristics of historical flow intensity measurements;
deriving a Gaussian mixture model having a probability density boundary; and
for each said mixture model, deriving a fitness score over a first predetermined period of time.

14. The method according to claim 13 further comprising:
testing each said mixture model for said first predetermined period of time using incoming flow intensity data to validate each said mixture model as a model based on whether each said fitness score is above a predetermined fitness threshold for said first predetermined period of time;
calculating a confidence score for each said mixture model by counting the number of times each said model fitness score is higher than said fitness threshold over a predetermined series of time windows;
using said model for fault detection so long as said confidence score for said model is above a predetermined confidence threshold; and
updating each said model confidence score over time.

15. The method according to claim 14 further comprising:
generating a residual for said mixture model, wherein said residual is the difference between a probability density of data points in a cluster and a probability density on a boumdary; comparing said residual to determine whether said residual is located within an ellipse defined by said model.

16. A method for detecting faults in a distributed transaction system comprising:
receiving data corresponding to flow intensity measured at a plurality of monitoring points in the distributed transaction system during normal operation of the system;
organizing said received flow intensity data to characterize a plurality of segments between every two monitoring points wherein said segments comprise at least one component; deriving a relationship for each said segment;
calculating a fitness score for each said relationship;
sequentially testing each said relationship for a predetermined period of time using flow intensity data to validate each said relationship as a model based on whether each said relationship fitness score for a model is above a predetermined fitness threshold for said predetermined period of time;
calculating a confidence score for each said model by counting the number of times each said model fitness score is higher than said fitness threshold;
using said model for fault detection so long as said confidence score for said model is above a predetermined confidence threshold;
updating each said model confidence score over time;
deriving a residual for each said model by tracking conformance between observed flow intensity measurements for each said segment and an output of each said model for that segment; using said confidence score for a model to evaluate how credible said residual is for that model when used for fault detection; and
correlating each said residual with its components to detect and isolate transaction system faults.

17. The method according to claim 16 wherein said models are linear regression models.

18. The method according to claim 16 wherein said models are Gaussian models.

19. A computer system comprising means adapted for carrying out the steps of a computerized method for detecting faults in a distributed transaction system comprising:
receiving data corresponding to flow intensity measured at a plurality of monitoring points in the distributed transaction system during normal operation of the system;
organizing said received flow intensity data to characterize a plurality of segments between every two monitoring points wherein said segments comprise at least one component; deriving a relationship for each said segment;
calculating a fitness score for each said relationship;
sequentially testing each said relationship for a predetermined period of time using flow intensity data to validate each said relationship as a model based on whether each said relationship fitness score for a model is above a predetermined fitness threshold for said predetermined period of time;
calculating a confidence score for each said model by counting the number of times each said model fitness score is higher than said fitness threshold;
using said model for fault detection so long as said confidence score for said model is above a predetermined confidence threshold;
updating each said model confidence score over time;
deriving a residual for each said model by tracking conformance between observed flow intensity measurements for each said segment and an output of each said model for that segment; using said confidence score for a model to evaluate how credible said residual is for that model when used for fault detection; and
correlating each said residual with its components to detect and isolate transaction system faults.

20. A computer program product comprising a machine-readable medium having computer-executable program instructions thereon including code means for causing a computer to perform a computerized method for detecting faults in a distributed transaction system comprising:
receiving data corresponding to flow intensity measured at a plurality of monitoring points in the distributed transaction system during normal operation of the system;
organizing said received flow intensity data to characterize a plurality of segments between every two monitoring points wherein said segments comprise at least one component; deriving a relationship for each said segment;
calculating a fitness score for each said relationship;
sequentially testing each said relationship for a predetermined period of time using flow intensity data to validate each said relationship as a model based on whether each said relationship fitness score for a model is above a predetermined fitness threshold for said predetermined period of time;
calculating a confidence score for each said model by counting the number of times each said model fitness score is higher than said fitness threshold;
using said model for fault detection so long as said confidence score for said model is above a predetermined confidence threshold;
updating each said model confidence score over time;
deriving a residual for each said model by tracking confoπnance between observed flow intensity measurements for each said, segment and an output of each said model for that segment; using said confidence score for a model to evaluate how credible said residual is for that model when used for fault detection; and
correlating each said residual with its components to detect and isolate transaction system faults.