2021、Autoformer¶
约 336 个字 17 张图片 预计阅读时间 2 分钟
- 模型结构图
主函数¶
序列分解模块¶
具体地操作通过箭头说明,方块里表示形状的变化
编码器模块¶
解码器模块¶
自相关机制¶
开始理论讲解¶
- 文章题目:Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
- 收录情况: 2021 NeurIPS
- 作者团队:清华软院吴海旭
- 开源地址:https://github.com/thuml/Autoformer
- Autoformer:Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
- https://openreview.net/pdf?id=I55UqU-M11y
- Autoformer: 一种基于自动相关机制的时序预测新架构
问题描述¶
序列分解预测模型
原文阅读¶
https://blog.csdn.net/sinat_37574187/article/details/144396724
本文聚焦的两个问题:
However, the forecasting task is extremely challenging under the long-term setting.
First, it is unreliable to discover the temporal dependencies directly from the long-term time series because the dependencies can be obscured by entangled temporal patterns.
Second, canonical Transformers with self-attention mechanisms are computationally prohibitive for long-term forecasting because of the quadratic complexity of sequence length.
(1)直接从长时间序列预测中发现时间模式 不太靠谱。因为距离太长了是其一,还有就是 时间模式错综复杂
(2)还是 Transformer 的计算复杂度太高了。现在一般采用稀疏的逐点计算,但是,损失信息了。
本文采用分解的概念。但是在之前的用处中,分解主要是预处理的步骤,因为未来数据的分解是观测不到的。
2025-03-17 12:04:36 2025-04-10 19:39:18