.png){kind=icon}
.png){kind=icon}
Return
A deep learning based multi-feature extraction approach for non-stationary time series
.png){kind=icon}
PRE
AI
S
DOI：10.1080/00949655.2026.2651217.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Majority of time series data is non-stationary, so the models are suffered from distribution shift problem for predicting the time series. Therefore, a deep learning-aided non-stationary time series forecasting approach is developed to handle various patterns and structural changes in data. Initially, the non-stationary time series data is collected from benchmark sites. Three different kinds of features are retrieved. The deep features are differentiated using Spatio-Temporal Attention based Sparse Autoencoder (STA-SA). All the three sets of features are utilized by the proposed Multi-scale Adaptive Residual-Echo State Network (MA-ResESN) model for performing the prediction process, which avoids poor prediction while using non-stationary time series data for performing tasks. The prediction performance is enhanced by fine tuning its parameter via Resilient Starfish Optimization Algorithm with Refined Random Parameter (RSOA-RRP). The prediction process of the developed deep learning-based approach is evaluated with existing non-stationary time series prediction models.
Keywords:
Non-stationary time series prediction
resilient starfish optimization algorithm with refined random parameter
multi-scale adaptive residual-echo state network
distributed principal component analysis
spatio-temporal attention based sparse autoencoder
multi-feature retrieval
Journal
J
IF:
1.2
Papers:
114
Citations:
4.1K
Organization
Cited Papers