A multiscale entropy gated DWTformer meteorological data time series prediction method and device

By improving the Transformer network through the DWTformer model with multi-scale entropy gating, and combining discrete wavelet decomposition and self-attention mechanism, the problem of difficulty in mining temporal dependencies in meteorological data time series forecasting is solved, and efficient forecasting of meteorological data is achieved.

CN117094431BActive Publication Date: 2026-05-29CHINA UNIV OF GEOSCIENCES (WUHAN) +2
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHINA UNIV OF GEOSCIENCES (WUHAN)
Filing Date
2023-08-01
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing meteorological data time series forecasting methods are difficult to effectively mine the potential time series dependencies of time series with highly nonlinear and non-stationary characteristics. In particular, in long-term forecasting tasks, the computational resource overhead is large and it is difficult to fully mine the potential time series dependencies.

Method used

A multi-scale entropy-gated DWTformer model is adopted, and the Transformer network is improved by combining time series decomposition methods. The time series is decomposed step by step through a multi-scale entropy-gated discrete wavelet decomposition module to construct a deep time series decomposition network. The time series features at different time scales are extracted by an exponential smoothing-based trend prediction module and a Wasserstein distance-based self-attention mechanism, respectively.

Benefits of technology

It achieves efficient forecasting of meteorological data, fully explores its potential time-series dependence characteristics, and improves forecasting performance, especially showing good forecasting accuracy in long-term forecasting tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117094431B_ABST
    Figure CN117094431B_ABST
Patent Text Reader

Abstract

The application provides a multiscale entropy gated DWTformer meteorological data time series prediction method and device, and belongs to the technical field of meteorological time series prediction. Based on the Transformer network, a multiscale entropy gated discrete wavelet time series decomposition module is designed to realize adaptive decomposition of periodic terms and trend terms to describe data change trends at different time scales, and a Wasserstein self-attention mechanism and an exponential smoothing prediction module are introduced to extract features at different frequency scales, fully excavate the time series dependence between production data, and effectively solve the problem of meteorological data prediction with nonlinear and non-stationary characteristics. The improved Transformer time series prediction model DWTformer proposed by the application performs multiscale time series feature extraction and prediction on meteorological data, better excavates the potential time series dependence of meteorological data, improves the prediction accuracy, and the experimental results show that the multiscale entropy gated DWTformer meteorological data time series prediction method has higher prediction accuracy than existing methods.
Need to check novelty before this filing date? Find Prior Art