GCformer combines a structured global convolutional branch for processing long input sequences with a local Transformer-based branch for capturing short, recent signals. Experiments demonstrate that GCformer outperforms state-of-the-art methods, reducing MSE error in multivariate time series benchmarks by 4.38% and model parameters by 61.92%. In particular, the global convolutional branch can serve as a plug-in block to enhance the performance of other models, with an average improvement of 31.93%, including various recently published Transformer-based models.

Figure 1. GCformer overall framework

Figure 2. Different parameterization methods of global convolution kernel

||

||

1. Install Python 3.6, PyTorch 1.11.0.
2. Download data. You can obtain all the six benchmarks from [FEDformer] or [Autoformer].
3. Train the model. We provide the experiment scripts of all benchmarks under the folder ./scripts/GCformer. For instance, you can reproduce the experiment result on illness dataset by:

bash ./scripts/GCformer/illness.sh

We appreciate the following github repos a lot for their valuable code base or datasets:

https://github.com/yuqinie98/PatchTST

https://github.com/MAZiqing/FEDformer

https://github.com/ctlllll/SGConv

https://github.com/thuml/Autoformer