Method and apparatus for generating time series data sets for predictive analysis

Patent metadata

Field	Value
Publication number	US-10185893-B2
Application number	US-201715440629-A
Country	US
Kind code	B2
Filing date	Feb 23, 2017
Priority date	Feb 29, 2016
Publication date	Jan 22, 2019
Grant date	Jan 22, 2019

Abstract

A computer-implemented method of generating, from time-series data, a time-series of data sets for predictive analysis, comprises dividing the time-series data into evenly-sized overlapping segments of data, generating an image representing data for each segment, using the time-series data to determine a trend associated with each image, and storing each of the generated images and its associated trend as a data set. In some embodiments of the method the image from each stored data set is transformed into numerical vectors through a feature extraction process using a pre-trained convolutional neural network. The numerical vectors are stored in association with the data set, and the data sets and associated numerical vectors are used to predict the trend for a new time-series image which has been generated from any time-series data.

First claim

What is claimed is:

1. A computer-implemented method of generating, from time-series data, a time-series of data sets for predictive analysis, which method comprises:
dividing the time-series data into evenly-sized overlapping segments of data;
for each segment, generating an image representing data in the segment producing generated images;
using the time-series data to determine a trend associated with each image;
storing each of the generated images and the trend associated with each image as a data set;
transforming each image from each stored data set into numerical vectors through a feature extraction process using a pre-trained convolutional neural network, and storing numerical vectors in association with the data set; and
using the data sets and associated numerical vectors to predict the trend for a new time-series image which has been generated from other time-series data.

2. A method as claimed in claim 1 , further comprising:
deriving a trend prediction model from the numerical vectors and trends for one of each stored data set and a sub-set of generated images, using a deep learning method;
predicting the trend for the new time-series image using the trend prediction model derived; and
outputting a predicted trend for the new time-series image.

3. A method as claimed in claim 1 , further comprising determining, for one of each generated image and a sub-set of generated images, a degree of similarity between generated image and each of other generated images, using the numerical vectors for the generated images.

4. A method as claimed in claim 3 , further comprising:
identifying a preset number k of the generated images which are a most similar to the new time-series image, where k is an integer;
using the trends associated with k identified images to determine an average trend; and
outputting the average trend as a predicted trend for the new time-series image.

5. A method as claimed in claim 1 , wherein each segment comprises contiguous first and second sub-segments of data, data in the first sub-segment being used to generate the image for the segment, and data in the second sub-segment being used to determine the trend associated with the image.

6. Data processing apparatus configured to generate, from time-series data, a time-series of data sets for predictive analysis, the apparatus comprising:
data preparation means for:
dividing the time-series data into evenly-sized overlapping segments of data;
for each segment, generating an image representing data in the segment producing generated images;
using the time-series data to determine a trend associated with each image;
storing each of the generated images and the trend associated with each image as a data set; and
transforming each image from each stored data set into numerical vectors through a feature extraction process using a pre-trained convolutional neural network, and store numerical vectors in association with the data set;
and trend prediction means for using the data sets and associated numerical vectors to predict the trend for a new time-series image which has been generated from other time-series data.

7. Apparatus as claimed in claim 6 , further comprising:
classifier training means for deriving a trend prediction model from the numerical vectors and trends for one of each stored data set and a sub-set of generated images, using a deep learning method;
wherein the trend prediction means predict the trend for the new time-series image using the trend prediction model derived and output a predicted trend for the new time-series image.

8. Apparatus as claimed in claim 6 , further comprising classifier training means for obtaining, in respect of one of each generated image and a sub-set of generated images, image similarity results by determining a degree of similarity between-the generated image and each of other generated images, using the numerical vectors for the generated images.

9. Apparatus as claimed in claim 8 , wherein the trend prediction means:
identify, using the image similarity results obtained by the classifier training means, a preset number k of the generated images which are a most similar to the new time-series image, where k is an integer;
use the trends associated with k identified images to determine an average trend; and
output the average trend as a predicted trend for the new time-series image.

10. Apparatus as claimed in claim 6 , wherein each segment comprises contiguous first and second sub-segments of data, and data preparation means use data in the first sub-segment to generate the image for the segment, and use data in the second sub-segment to determine the trend associated with the image.

11. A non-transitory computer-readable storage medium including a computer program which, when run on a computer, causes the computer to carry out the method of claim 1 .

12. Data processing apparatus as claimed in claim 6 , wherein the data preparation means comprises a computing device.

Assignees

• Fujitsu Ltd

Inventors

• Townsend Joseph
• Mendes Rodrigues Eduarda

Classifications

• G06V10/82 Primary

  using neural networks · CPC title

• G06F2218/16

  by matching signal segments · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06F18/22

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