Multivariable input and multivariable output time series prediction method and system

A technology for outputting time and sequence prediction, applied in prediction, instrument, biological neural network model, etc., can solve the problem of inability to take into account the covariance of long-term dependencies and related factors at the same time, and achieve the effect of improving perception ability and prediction accuracy.

Pending Publication Date: 2021-03-26
ENJOYOR COMPANY LIMITED
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, limited by the limit of gradient transmission, the above methods cannot take into account the long-term dependence and the covariance of associated factors at the same time

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Multivariable input and multivariable output time series prediction method and system
  • Multivariable input and multivariable output time series prediction method and system
  • Multivariable input and multivariable output time series prediction method and system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0052] see figure 1 , this embodiment provides a multivariate input and multivariate output time series forecasting method, the specific steps are as follows:

[0053] S1. Prepare multivariate time series data X; multivariate time series data X includes at least one time variable data g t and at least one non-time variable data y t and in chronological order, i.e. x t ={g t ,y t}, multivariate time series data with X t ={x 1 , x 2 ,...,x t}, where t represents the last moment of the sequence, and the data at each moment is n-dimensional data, that is, there are n variables. such as x t ={xx 1 , xx 2 ,...,xx n} t . x t Among the n variables of , there are several variables that only change with time, such as year, month, day information, whether there are holidays, etc. These variables are recorded as g t , and the remaining variables are recorded as y as non-time variables t ,y t The dimension of is denoted as m.

[0054] Examples of multivariate time-series...

Embodiment 2

[0102] see figure 2 , realizing the multivariate input and multivariate output time series prediction system described in Embodiment 1, including

[0103] Convolutional network for extracting time series trend data R of multivariate time series data X;

[0104] The long-period encoding and decoding module includes a long-period encoder and a long-period decoder, wherein the long-period encoder is a self-attention layer, and the long-period decoder includes a self-attention layer and a self-attention decoding layer. Perform correlation calculation on the time series trend data R to obtain the time series trend correlation vector Corr and variable co-correlation vector Corr D ;

[0105] Regression prediction module, used for correlation vector Corr, time series trend data R of historical data and co-correlation vector Corr D Calculate the estimated value of the forecast time series variable Y;

[0106] The model weight training module is used to compare the system output va...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention provides a multivariable input and multivariable output time series prediction method and system, and the method comprises the following specific steps: S1, preparing multivariable timeseries data X; S2, performing sequential calculation on the multivariable sequential data X one by one by using a convolutional network to obtain sequential trend data R; S3, constructing a long-period encoding and decoding model based on an auto-attention model, and calculating a correlation vector Corr between the current time trend and the historical time trend and a covariant relationship CorrD between to-be-predicted time sequence variables; and S4, calculating an estimated value of the prediction time sequence variable Y by utilizing the correlation vector Corr, the cocorrelation vectorCorrD and the time sequence trend data R of the historical data.

Description

technical field [0001] The invention belongs to the field of time series forecasting, and relates to a multivariate input and multivariate output time series forecasting method and system. Background technique [0002] Time series forecasting can be divided into three cases: {univariate input, univariate output}, {multivariate input, univariate output} and {multivariate input, multivariate output}. The first case is the simplest and only considers historical values The influence on the future value, regardless of other influencing factors, the prediction accuracy is low due to the lack of reference information; the second case considers a variety of influencing factors, and the prediction accuracy is improved to a certain extent, but only for single variable output However, there are many variables to be predicted in the actual application scenario, so the scope of the application scenario is relatively small; the third situation is common in real life, such as predicting th...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06Q40/04G06Q10/04G06Q10/06G06N3/04
CPCG06Q40/04G06Q10/04G06Q10/067G06N3/045
Inventor 丁锴李建元陈涛
Owner ENJOYOR COMPANY LIMITED
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap