Method for characterization, detection and prediction for target events

• US 7,292,960 B1
• Filed: 06/30/2006
• Issued: 11/06/2007
• Est. Priority Date: 06/30/2006
• Status: Active Grant

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: GM Global Technology Operations Incorporated (General Motors Company)
• Inventors: Barajas, Leandro G., Srinivasa, Narayan, Jiang, Qin
• Primary Examiner(s): Cosimano; Edward R
• Application Number:
  US11/427,825
• Time in Patent Office: 494 Days
• Field of Search: 340/500, 340/511, 340/679, 340/680, 700/90, 700/108, 700/110, 702/33, 702/34, 702/35, 702/127, 702/179, 702/180, 702/182, 702/183, 702/184, 702/185, 702/188
• US Class Current: 702/185
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling

Abstract

A method for characterizing, detecting and predicting an event of interest, a target event, based on temporal patterns useful for predicting a probable occurrence of the target event is disclosed. Measurable events and their features are defined and quantized into event classes. Temporal series of the event classes are analyzed, and preliminary prediction rules established by analyzing temporal patterns of the event classes that precede an occurrence of the target event using a sliding time window. The quality of the preliminary prediction rules is evaluated and parameters thereof are optimized by using a defined fitness function, thereby defining finalized prediction rules. The finalized prediction rules are then made available for application on temporal series of the event classes to forecast a probable occurrence of the target event.

Claims

1. A method for characterizing, detecting and predicting an event of interest, a target event, based on temporal patterns useful for predicting a probable occurrence of the target event, the method comprising:

defining measurable events and their features and quantizing them into event classes;

analyzing temporal series of the event classes, and establishing preliminary prediction rules by analyzing temporal patterns of the event classes that precede an occurrence of the target event using a sliding time window;

evaluating the quality of and optimizing parameters of the preliminary prediction rules using a defined fitness function, thereby defining finalized prediction rules;

making available the finalized prediction rules for application on a temporal series of the event classes to forecast a probable occurrence of the target event; and

applying the finalized prediction rules on the temporal series of the event classes to forecast a probable occurrence of the target event.

2. The method of claim 1, further comprising:

agglomerating the quantized measurable events into event classes.

3. The method of claim 1, wherein the target event has a low frequency of occurrence, but a high impact upon its occurrence, relative to all of the other event classes.

4. The method of claim 1, wherein the event classes include both the target event and non-target events, and only the non-target events are used to define the finalized prediction rules.

5. The method of claim 1, wherein the event classes result from more than one data set, thereby enabling multivariable predictions.

6. The method of claim 1, wherein the quantizing the defined measurable events into event classes comprises:

merging neighboring data points defining the measurable events until a defined number of event classes are established.

7. The method of claim 1, wherein the defined fitness function is defined to vary as a function of a confusion matrix.

8. The method of claim 1, further comprising:

in response to the forecasted probable occurrence of the target event, enabling a machine health management system to take action to prevent a future occurrence of the target event or to prepare for a resolution if the target event were to occur, thereby transforming the measurable events into an identifiable actionable event.

9. The method of claim 8, wherein:

the actionable event comprises preventative maintenance of a machine, a device or a network.

10. The method of claim 1, wherein related events are defined as the event classes in the sliding window that precede an occurrence of the target event, and the temporal order of the related events is not considered in defining the finalized prediction rules.

11. The method of claim 10, wherein the evaluating the quality of the preliminary prediction rules comprises:

using the sliding window, analyzing the temporal series of the event classes to find an analysis time window containing a set of related events;

determining whether a target event is present in the next sequential time window, and based OD this information, computing four values of a two-by-two confusion matrix; and

using the four values to compute a fitness score based on the fitness function.

12. The method of claim 10, wherein the establishing the preliminary prediction rules comprises:

analyzing the temporal series of the event classes to find all analysis time windows preceding the target event;

extracting sets of the related events from the analysis time windows;

using the sets of related events, constructing an event space containing all possible event patterns generated by the sets of related events;

using the event space, finding all frequent event sets, the frequent event sets being defined as an event set with a frequency of occurrence larger than a pre-defined minimum;

using the event space, finding confident event sets, the confident event sets being defined as an event set with a confidence measure for predicting the occurrence of the target event larger than a pre-defined minimum; and

defining target event predictors as the intersection of both the frequent event sets and the confident event sets.

13. The method of claim 12, wherein the optimizing comprises:

defining a set of system parameters relating to the event classes, and encoding the system parameters to define an encoded chromosome; and

employing a genetic algorithm to assess the value of the system parameters within the encoded chromosome by applying a first valuation method and a second valuation method.

14. The method of claim 13, wherein:

the first valuation method comprises evaluating a frequency score and a confidence score associated with each target event predictor.

15. The method of claim 13, wherein:

the second valuation method comprises evaluating a confusion matrix associated with each target event predictor.

16. The method of claim 12, wherein the optimizing comprises:

defining a set of system parameters relating to the event classes, encoding the system parameters to define an encoded chromosome, and defining the fitness function such that it varies as a function of the length of the encoded chromosome.

17. The method of claim 16, wherein:

the encoded chromosome comprises a binary string having a plurality of binary string segments, each segment defining a unique one of the system parameters, and several chromosomes are capable of being encoded together to solve multivariable input problems.

18. The method of claim 17, wherein:

one of the segments defines one of;

a size of the sliding window;

a bin size of a histogram of the measurable events;

a number of the event classes; and

, the target event class.

19. The method of claim 16, wherein the optimizing comprises:

employing a genetic algorithm to assess the value of the chromosome by calculating a set of scores that are a function of a confusion matrix, process parameters, and algorithm parameters.

20. The method of claim 19, wherein the set of scores comprises at least one of a precision score, an accuracy score and a false positive rate score.

21. The method of claim 20, wherein the optimizing comprises:

employing a genetic algorithm to select values for the system parameters for the encoded chromosome through optimization of the fitness function.

22. The method of claim 21, wherein the fitness function is defined as:

fit=S₁*S₂*(1−

S₃)S₄*len;

where;

S₁ is the precision score;

S₂ is the accuracy score;

S₃ is false positive rate score;

len is the length of the chromosome; and

S₄ is defined as;

S₄=f+cf,

where;

f is a frequency score; and

cf is a confidence score;

wherein each target event predictor has an associated frequency score based its frequency of occurrence, and an associated confidence score based on its confidence measure.

23. A computer usable medium having computer readable program code for characterizing, detecting and predicting an event of interest, a target event, based on temporal patterns useful for predicting a probable occurrence of the target event, the computer readable program code executable by a computer for facilitating:

defining measurable events and their features and quantizing them into event classes;

analyzing temporal series of the event classes, and establishing preliminary prediction rules by analyzing temporal patterns of the event classes that precede an occurrence of the target event using a sliding time window;

evaluating the quality of and optimizing parameters of the preliminary prediction rules using a defined fitness function, thereby defining finalized prediction rules; and

making available the finalized prediction rules for application on a temporal series of the event classes to forecast a probable occurrence of the target event.