1.School of Computer and Communication Engineering,Northeastern University at Qinhuangdao,Qinhuangdao 066003,China
2.Puyang Power Supply Company,State Grid Henan Electric Power Company,Puyang 457000,China. Corresponding author: GUO Rui-xin,E-mail: 1040980690@qq. com
Existing time series forecasting models overlook the effective fusion of variable information and temporal information embedded in raw data during feature extraction, which results in the failure to fully utilize the information of raw data. To address this issue, a gated-driven time series forecasting method based on temporal and variable information fusion (GFTV) was proposed. This model leveraged trend-seasonal decomposition techniques and employed a Transformer architecture to extract both cross-temporal and cross-variable features from the raw data. Subsequently, a gating fusion mechanism based on gating units was used to fuse the cross-temporal and cross-variable features, enhancing the model’s ability to capture data information. The model achieved optimal performance on both the ETTh1 and Weather datasets, and ablation studies confirmed the effectiveness of the modules contained in GFTV, providing a novel approach for dynamically fusing temporal and variable information in the field of time series forecasting.
VaswaniA, ShazeerN, ParmarN, et al. Attention is all you need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach,2017: 6000-6010.
[2]
ZengA L, ChenM X, ZhangL, et al. Are transformers effective for time series forecasting? [EB/OL]. (2023-06-26)[2024-07-23].
[3]
ZhouH Y, ZhangS H, PengJ Q, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(12): 11106-11115.
[4]
WuH X, XuJ H, WangJ M, et al. Autoformer: decomposition transformers with auto-correlation for long-term series forecasting[J]. Advances in neural information processing systems,2021, 34:22419-22430.
[5]
ZhouT, MaZ Q, WenQ S, et al. FEDformer: frequency enhanced decomposed transformer for long-term series forecasting[C] // International conference on machine learning. Baltimore.2022:17-23.
[6]
NieY Q, NguyenN H, SinthongP, et al. A time series is worth 64 words: long-term forecasting with transformers [EB/OL]. (2022-11-27)[2024-07-23].
[7]
GaoJ X, HuW B, ChenY T. Client: cross-variable linear integrated enhanced transformer for multivariate long-term time series forecasting[EB/OL]. (2023-05-23)[2024-09-23].
[8]
LiuY, HuT G, ZhangH R, et al. iTransformer: inverted transformers are effective for time series forecasting[EB/OL]. (2023-10-10)[2024-09-23].
[9]
LiS Y, JinX Y, XuanY, et al. Enhancing the locality and breaking the memory bottleneck of transformer on time series forecasting[C]//33rd Conference on Neural Information Processing Systems(NeurIPS 2019. Vancouver. 2019: 1-11.
[10]
LiuY, WuH X, WangJ M, et al. Non-stationary transformers: exploring the stationarity in time series forecasting[EB/OL]. (2023-11-24)[2024-07-23].
[11]
WuH X, HuT G, LiuY, et al. TimesNet: temporal 2D-variation modeling for general time series analysis[EB/OL]. (2023-04-12)[2024-09-23].