论文笔记:Adjusting for Autocorrelated Errors in Neural Networks for Time Series

2021 NIPS

  • 原来的时间序列预测任务是根据\{X_{t-L},X_{t-L+1},\cdots,X_t\}预测X_{t+1}
  • 论文提出用一阶自回归误差{X_{t-L}-\rho X_{t-L-1},X_{t-L+1}-\rho X_{t-L},\cdots,X_t-\rho X_{t-1}}预测X_{t+1}-\rho X_t
    • 一阶差分,类似于ResNet的残差思路?
    • X_{t+1}-\rho X_t为pred,最终的预测结果X_{t+1}=pred+\rho X_t

论文笔记:Adjusting for Autocorrelated Errors in Neural Networks for Time Series_第1张图片

 

你可能感兴趣的:(论文笔记,论文阅读)