System, device and method for automatic anomaly detection

• US 7,519,860 B2
• Filed: 06/06/2001
• Issued: 04/14/2009
• Est. Priority Date: 09/11/2000
• Status: Expired due to Fees

• Current Assignee: Nokia Corporation
• Original Assignee: Nokia Corporation
• Inventors: Hoglund, Albert, Hatonen, Kimmo, Berg, Jyrki, Torikka, Sampo, Kulmala, Kari, Miettinen, Markus
• Primary Examiner(s): Puente; Emerson C
• Application Number:
  US10/363,745
• Publication Number:
  US 20040039968A1
• Time in Patent Office: 2,869 Days
• Field of Search: 714/26, 726/23
• US Class Current: 714/26
• CPC Class Codes: G06F 18/2433 Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

A method and system for monitoring the behavior of at least one observable object, e.g. a network element, of a network, wherein at least one parameter of the observable object is repeatedly detected. An actually detected parameter is input to a learning process and to an analyzing process, wherein the learning process forms a reference, based on at least two detected parameter values, for describing the behavior of the observable object. The analyzing process compares the input parameter and the reference for detecting an anomal behavior. The parameter preferably is a vector which includes several values describing properties or functioning of the observable object, and is formed based on events and/or reports from the object.

Claims

1. A method, comprising:

repeatedly detecting at least one parameter of at least one observable object of a network;

checking the at least one parameter with regard to fulfilling predetermined criteria; and

forming a vector based on the at least one parameter depending on whether the at least one parameter fulfills the predetermined criteria,

wherein the vector comprises several values which describe properties or functioning of the observable object, and the formed vector is evaluated to monitor a behavior of the observable object, and

wherein a number of parameters that fulfill the predetermined criteria during an observation period, is counted, to form key performance indicator (KPI) values which form part of the vector, each parameter having its own criteria.

2. The method according to claim 1, further comprising:

inputting the formed vector into a learning process and an analyzing process,

wherein the learning process is configured to form a reference, based on the input vector and a previous value of the reference or at least one previously input vector, to describe the behavior of the observable object, and

wherein the analyzing process is configured to compare the input vector and the reference to detect anomalous behavior.

3. The method according to claim 2, wherein the reference formed by the learning process is a profile generated from at least two vectors.

4. The method according to claim 2, wherein the learning process comprises a self organizing map (SOM).

5. The method according to claim 2, wherein the learning process comprises a K-means algorithm.

6. The method according to claim 2, further comprising:

using the vector as input data to an algorithm which learns the normal functioning of the network, and forms the reference as a profile describing the normal functioning of the network, the profile consisting of nodes or cluster centroids,

finding best mapping nodes or cluster centroids of the profile for a new incoming vector;

counting distance between the new incoming vector and the best mapping nodes or cluster centroids;

counting a distribution of distances;

checking from the distribution whether a distance of the new incoming vector represents a probability which is less than a predetermined probability set-up value; and

generating an alarm, if the check result is positive.

7. The method according to claim 1, wherein, when a real time traffic (RTT) report value fulfills the predetermined criteria, value of one key performance indicator (KPI) is increased by one.

8. The method according to claim 7, wherein a real time traffic (RTT) report containing values of the number of parameters comprises fields defining

end cause code,

a length of call,

sender identification,

receiver identification,

location information,

used network resources, and/or

used services.

9. The method according to claim 1, wherein key performance indicator (KPI) value is a function of at least one predetermined real time traffic (RTT) report field values of such RTT reports that fulfill the predetermined criteria.

10. The method according to claim 1, wherein key performance indicator (KPI) value is a function of at least one other KPI.

11. The method according to claim 1, wherein the predetermined criteria are checked by comparing fields of a real time traffic (RTT) report to predetermined field thresholds.

12. The method according to claim 1, wherein the predetermined criteria and the observation period are user definable.

13. The method according to claim 1, wherein the vector is formed based on detected values in real time traffic (RTT) reports.

14. A system, comprising:

a detector configured to detect at least one parameter of at least one observable object of a network;

a checking circuit configured to check the at least one parameter with regard to fulfilling predetermined criteria;

a forming circuit configured to form a vector based on the at least one parameter depending on whether the at least one parameter fulfills the predetermined criteria, wherein the vector comprises several values which describe properties or functioning of the observable object; and

an evaluating circuit configured to evaluate the formed vector to monitor a behavior of the observable object,

wherein a number of parameters that fulfill the predetermined criteria during an observation period, is counted, to form key performance indicator (KIN) values which form part of the vector, each parameter having its own criteria.

15. An apparatus, comprising:

a detector configured to repeatedly detecting at least one parameter of at least one observable object of a network;

a checking circuit configured to check the at least one parameter with regard to fulfilling predetermined criteria; and

a forming circuit configured to form a vector based on the at least one parameter depending on whether the at least one parameter fulfills the predetermined criteria, wherein the vector comprises several values which describe properties or functioning of the observable object; and

an evaluating circuit configured to evaluate the formed vector to monitor a behavior of the observable object,

wherein a number of parameters that fulfill the predetermined criteria during an observation period, is counted, to form key performance indicator (KPI) values which form part of the vector, each parameter having its own criteria.

16. The apparatus according to claim 15, further comprising:

a learning circuit configured to receive an actually calculated vector and to form a reference, based on at least two vectors, to describe the behavior of the observable object; and

an analyzing circuit configured to analyze the actually calculated vector and to compare the received vector and the reference to detect anomalous behavior.

17. The apparatus according to claim 16, wherein the learning circuit includes a self organizing map (SOM) or K-means algorithm.

18. The apparatus according to claim 15, wherein the parameter is formed based on detected real time traffic (RTT) report field values.

19. The apparatus according to claim 15, further comprising:

an inputting circuit configured to input the vector as input data to an algorithm which learns the normal functioning of the network, and forms, as a reference, a profile describing the normal functioning of the network, the profile consisting of nodes or cluster centroids;

a locating circuit configured to find best mapping nodes or cluster centroids of the profile for a new incoming vector;

a counting circuit configured to count distance between the new incoming vector and the best mapping nodes or cluster centroids;

a counting circuit configured to count a distribution of distances;

a checking circuit configured to check whether a distance of an incoming vector has a probability value less than a predetermined probability set-up value; and

a generating circuit configured to generate an alarm, if the check result is positive.

20. An apparatus, comprising:

detecting means for detecting at least one parameter of at least one observable object of a network;

checking means for checking the at least one parameter with regard to fulfilling predetermined criteria;

forming means for forming a vector based on the at least one parameter depending on the result of the checking means, wherein the vector comprises several values which describe properties or functioning of the observable object; and

evaluating means for evaluating the formed vector for monitoring the behavior of the observable objects,

wherein a number of parameters that fulfill the predetermined criteria daring an observation period, is counted, to form key performance indicator (KPI) values which form part of the vector, each parameter having its own criteria.

21. A computer program, embodied on a computer readable medium, configured to control a processor to implement a method, the method comprising:

repeatedly detecting at least one parameter of at least one observable object of a network;

checking the at least one parameter with regard to fulfilling predetermined criteria; and

forming a vector based on the at least one parameter depending on whether the at least one parameter fulfills the predetermined criteria,

wherein the vector comprises several values which describe properties or functioning of the observable object, and the formed vector is evaluated to monitor a behavior of the observable object, and

wherein a number of parameters that fulfill the predetermined criteria during an observation period, is counted, to form key performance indicator (KPI) values which form pall of the vector, each parameter having its own, criteria.