计算机科学 ›› 2026, Vol. 53 ›› Issue (2): 180-186.doi: 10.11896/jsjkx.250100113
潘建1,2, 汪绪豪2
PAN Jian1,2, WANG Xuhao2
摘要： 目前,在时间序列预测任务的研究中,基于Transformer的模型主要关注从时序数据中提取全局性和局部性特征,并通过改进注意力机制以降低模型的复杂度。然而,现有方法往往忽略了时间序列在多个尺度上展现出的不同粒度特征。针对这一问题,提出了一种融合多尺度特征和注意力机制的时间序列预测模型——MTSformer。首先,通过对原始序列进行下采样,得到多个尺度的子序列,使模型能够融合多个尺度的特征信息,从而增强泛化能力;其次,使用多预测头代替传统的解码器,在提升预测速度的同时降低模型的复杂度;最后,在5个基准数据集上进行了实验,结果显示,与现有的方法相比,MTSformer模型在时间序列预测上的MSE平均降低了24.51%,MAE平均降低了17.84%。
中图分类号:
| [1]MAO Y H,SUN C C,XU L Y,et al.A Review of Time Series Forecasting Methods Based on Deep Learning[J].Microelectronics & Computer,2023,40(4):8-17. [2]CHENG W H,CHE W G.Research on Financial Time SeriesForecasting Algorithm Based on Secondary Decomposition and LSTM[J].Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition),2022,34(4):638-645. [3]KAN G Y,YANG J.Research on Meteorological Time Series Forecasting Based on Wavelet Transform and LSTM Hybrid Model[J].Computer Science and Application,2022,12(3):682-689. [4]XU S,LIU D D.Power Load Forecasting Based on Time Series Combination Model[J].Electronic Design Engineering,2023,31(23):1-6. [5]VASWANI A,SHAZEER N,PARMAR N,et al.Attention isall you need[C]//Proceedings of the 31st InternationalConfe-rence on Neural Information Processing Systems.2017:6000-6010. [6]ZAREMBA W,ILYA S,ORIOL V.Recurrent neural network regularization[J].arXiv:1409.2329,2014. [7]CHUNG J,GULCEHRE C,CHO K H,et al.Empirical evalu-ation of gated recurrent neural networks on sequence modeling[J].arXiv:1412.3555,2014. [8]HOCHREITER S,SCHMIDHUBER J.Long short-termmemory[J].Neural Computation,1997,9(8):1735-1780. [9]LEA C,FLYNN M D,VIDAL R,et al.Temporal convolutional networks for action segmentation and detection[C]//Procee-dings of the IEEE Conference on Computer Vision and Pattern Recognition.2017:156-165. [10]WU H,HU T,LIU Y,et al.Timesnet:Temporal 2d-variation modeling for general time series analysis[C]//The Eleventh International Conference on Learning Representations.2022. [11]ZHOU H,ZHANG S,PENG J,et al.Informer:Beyond efficient transformer for long sequence time-series forecasting[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2021:11106-11115. [12]WU H,XU J,WANG J,et al.Autoformer:Decomposition transformers with auto-correlation for long-term series forecasting[J].Advances in Neural Information Processing Systems,2021,34:22419-22430. [13]KITAEV N,KAISER Ł,LEVSKAYA A.Reformer:The efficient transformer[J].arXiv:2001.04451,2020. [14]ZHOU T,MA Z,WEN Q,et al.Fedformer:Frequency enhanced decomposed transformer for long-term series forecasting[C]//International Conference on Machine Learning.PMLR,2022:27268-27286. [15]NIE Y,NGUYEN N H,SINTHONG P,et al.A time series is worth 64 words:Long-term forecasting with transformers[J].arXiv:2211.14730,2022. [16]CHEN P,ZHANG Y,CHENG Y,et al.Pathformer:Multi-scale transformers with adaptive pathways for time series forecasting[J].arXiv:2402.05956,2024. [17]ZENG A L,CHEN M X,ZHANG L,et al.Are transformers effective for time series forecasting?[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2023. [18]ASUNCION A,NEWMAN D.UCI machine learning repository[EB/OL].https://github.com/uci-ml-repo. [19]CHOLLET F.Deep learning with Python[M]//Manning Publications.2021:45-52. [20]ZHAO L,GKOUNTOUNA O,PFOSER D.Spatial auto-regressive dependency interpretable learning based on spatial topological constraints[J].ACM Transactions on Spatial Algorithms and Systems,2019,5(3):1-28. |
|
||
首页