Prediction of an event affecting a physical system

Patent metadata

Field	Value
Publication number	US-2019197424-A1
Application number	US-201816232650-A
Country	US
Kind code	A1
Filing date	Dec 26, 2018
Priority date	Dec 25, 2017
Publication date	Jun 27, 2019
Grant date	—

Abstract

The disclosure notably relates to a computer-implemented method for predicting new occurrences of an event of a physical system. The method comprises providing a first set of past events of the physical system, each past event comprising several attributes, providing a signature for each past event of the first set, providing a new event comprising several attributes, computing a signature of the new event, computing a similarity measure between the signature of the new event and each signature of each past event of the first set, determining the past events closest to the new event according to the similarity measures thereby forming a second set of past events, computing a score of relevance for each attribute of the second set, providing a set of attributes by selecting the attributes having the greater scores of relevance.

First claim

1 . A computer-implemented method for predicting new occurrences of an event of a physical system, comprising:
obtaining a first set of past events of the physical system, each past event comprising several attributes; obtaining a signature for each past event of the first set; obtaining a new event comprising several attributes; computing a signature of the new event; computing a similarity measure between the signature of the new event and each signature of each past event of the first set; determining the past events closest to the new event according to the similarity measures thereby forming a second set of past events; computing a score of relevance for each attribute of the second set; and generating a set of attributes by selecting the attributes having the greater scores of relevance.

2 . The computer-implemented method of claim 1 , wherein the attributes of the second set for which the scores of relevance are computed are present in both the first and second sets.

3 . The computer-implemented method of claim 2 , wherein the score of relevance of an attribute is computed by comparing distribution probabilities of values of the attribute on the second set with distribution probabilities of values of the attribute on the first set.

4 . The computer-implemented method of claim 1 , wherein obtaining a signature for each past event of the first set and for the new event comprises obtaining a numerical vector for each past event of the first set and for the new event.

5 . The computer-implemented method of claim 4 , wherein obtaining a signature includes:
training a machine learning model with a third set of past events; and applying, on each past event of the first set and on the new event, the machine learning model.

6 . The computer-implemented method of claim 5 , wherein the trained model is a context sensitive auto-encoder.

7 . The computer-implemented method of claim 1 , wherein a similarity metric used for determining the past events closest to the new event is one among:
cosine similarity; Euclidian distance; and inverse Euclidian distance.

8 . The computer-implemented method of claim 1 , wherein an attribute comprises at least one structured attribute.

9 . The computer-implemented method of claim 8 , wherein an attribute further comprises at least one unstructured attribute.

10 . The computer-implemented method of claim 8 , wherein the score of relevance is computed with values of structured attributes only.

11 . The computer-implemented method of claim 1 , further comprising, after forming a second set of past events:
computing at least one subset of past events of the second set; and wherein computing the score of relevance further includes: computing the score of relevance for each attribute of in the said at least one subset of past events of the second set.

12 . The computer-implemented method of claim 1 , wherein the past events of the first set and the new event are described by the same attributes.

13 . The computer-implemented method of claim 1 , wherein the new event and the signature of the new event are stored with the past events and the signatures of the past events thereby becoming a past event.

14 . A non-transitory computer readable medium having stored thereon a computer program comprising instructions for performing a method for predicting new occurrences of an event of a physical system, the method comprising:
obtaining a first set of past events of the physical system, each past event comprising several attributes; obtaining a signature for each past event of the first set; obtaining a new event comprising several attributes; computing a signature of the new event; computing a similarity measure between the signature of the new event and each signature of each past event of the first set; determining the past events closest to the new event according to the similarity measures thereby forming a second set of past events; computing a score of relevance for each attribute of the second set; and generating a set of attributes by selecting the attributes having the greater scores of relevance.

15 . A system comprising:
a processor coupled to a memory, the memory having recorded thereon a computer program for predicting new occurrences of an event of a physical system that when executed by the processor causes the processor to be configured to: obtain a first set of past events of the physical system, each past event comprising several attributes; obtain a signature for each past event of the first set; obtain a new event comprising several attributes; compute a signature of the new event; compute a similarity measure between the signature of the new event and each signature of each past event of the first set; determine the past events closest to the new event according to the similarity measures thereby forming a second set of past events; compute a score of relevance for each attribute of the second set; and generate a set of attributes by selecting the attributes having the greater scores of relevance.

16 . The computer-implemented method of claim 2 , wherein providing a signature for each past event of the first set and for the new event comprises providing a numerical vector for each past event of the first set and for the new event.

17 . The computer-implemented method of claim 3 , wherein providing a signature for each past event of the first set and for the new event comprises providing a numerical vector for each past event of the first set and for the new event.

18 . The computer-implemented method of claim 2 , wherein a similarity metric used for determining the past events closest to the new event is one among:
cosine similarity; Euclidian distance; and inverse Euclidian distance.

19 . The computer-implemented method of claim 3 , wherein a similarity metric used for determining the past events closest to the new event is one among:
cosine similarity; Euclidian distance; and inverse Euclidian distance.

20 . The computer-implemented method of claim 4 , wherein a similarity metric used for determining the past events closest to the new event is one among:
cosine similarity; Euclidian distance; and inverse Euclidian distance.

Assignees

• Dassault Systemes

Inventors

• Grehant Xavier

Classifications

• G06F16/2462 Primary

  Approximate or statistical queries · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06F18/241

  relating to the classification model, e.g. parametric or non-parametric approaches · CPC title

• G06F18/22 Primary

  Matching criteria, e.g. proximity measures · CPC title