.png){kind=icon}
Return
AICNet: Adaptive Interactive Convolution Network for multivariate time series forecasting
.png){kind=icon}
PRE
AI
DOI：10.1016/j.measurement.2026.122717.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Time series forecasting plays a crucial role in fields such as meteorology, traffic management, and economic planning. The inherent long-term and short-term dependencies present unique challenges for predictive applications. Although Transformer-based models excel at capturing long-term dependencies, they still face limitations in noise sensitivity, computational efficiency, and overfitting issues on small datasets. To overcome these difficulties, An Adaptive Interactive Convolutional Network for multivariate time series forecasting (AICNet) has been designed, aiming to serve as a comprehensive solution for various multivariate time series tasks. Specifically, this work introduces a weighted adaptive high-frequency masking spectral block, which utilizes Fourier transforms to enhance feature representation. This block dynamically adjusts the weights of frequency components through an adaptive high-frequency masking mechanism, automatically selecting the frequency information most relevant to the forecasting task. This approach not only effectively captures extended and short-term relationships but also reduces the impact of noise on model performance through adaptive masking. This work also designed an interactive batch convolution block to enhance AICNet's ability to decode complex temporal patterns, improving its robustness across different datasets. Experimental results demonstrate that AICNet outperforms existing state-of-the-art models across various forecasting tasks, showcasing its resilience and adaptability across different data scale.
Keywords:
Time series forecasting
Interactive convolution
Fourier transform
Journal
IF:
5.6
Papers:
2.1W
Citations:
5.4W
Organization
Cited Papers
No cited papers available