Title:
MULTIVARIABLE TIME SEQUENCE PROCESSING METHOD AND APPARATUS, DEVICE AND MEDIUM
Document Type and Number:
WIPO Patent Application WO/2023/221701
Kind Code:
A1
Abstract:
Disclosed in the present application is a multivariable time sequence processing method, comprising: obtaining a time sequence set for mining a causal relationship between variables in a multivariable time sequence, the time sequence set comprising a plurality of first time sequence fragments, wherein each first time sequence fragment has the same length and is a multivariable time sequence; for any first time sequence fragment, inputting the first time sequence fragment into a graph neural network, and predicting a multivariable parameter value corresponding to the first moment, wherein the first moment is the next moment adjacent to the latest moment in the first time sequence fragment; determining an optimization function on the basis of multivariable parameter values corresponding to a plurality of first moments and multivariable sequence tags (true values) corresponding to the plurality of first moments, the optimization function comprising a loss function and a causal matrix of the graph neural network; determining a value of each parameter in the causal matrix by taking minimization of the optimization function as a target; and determining the causal relationship among a plurality of variables according to the value of each parameter.
Inventors:
SUN YANG (CN)
XIE YIFAN (CN)
LIU KANG (CN)
LUO XUAN (CN)
XIE YIFAN (CN)
LIU KANG (CN)
LUO XUAN (CN)
Application Number:
PCT/CN2023/088339
Publication Date:
November 23, 2023
Filing Date:
April 14, 2023
Export Citation:
Assignee:
BEIJING VOLCANO ENGINE TECH CO LTD (CN)
International Classes:
G06T19/00
Foreign References:
CN114925811A | 2022-08-19 | |||
CN114169394A | 2022-03-11 | |||
CN112101480A | 2020-12-18 | |||
CN114372504A | 2022-04-19 | |||
US20190034497A1 | 2019-01-31 |
Attorney, Agent or Firm:
SINOTALEN ATTORNEYS AT LAW (CN)
Download PDF: