加载中...
本论文提出 BASTION:一种基于贝叶斯的自适应趋势与多重季节性分解框架,结合惩罚化非参数回归与全局-局部(horseshoe)收缩先验以在存在突变、异常值和时变波动时实现稳健且局部自适应的趋势/季节性估计;此外,给出分解可辨识性的严格条件并通过模拟与实证(航空客流、纽约电力负荷)验证 BASTION 在估计精度与置信区间校准上的优势,且实现可线性扩展的 Gibbs 采样算法并以 R 包形式开源
| DGP / Component | TBATS | MSTL | STR | BASTION |
|---|---|---|---|---|
| DGP1 - Signal | 13.878 | 11.760 | 10.829 | 0.376 |
| DGP1 - Trend | 3.624 | 13.227 | 13.785 | 0.581 |
| DGP1 - Season | 11.672 | 2.712 | 2.575 | 0.536 |
| DGP2 - Signal | 1.411 | 1.214 | 0.7020 | 0.3922 |
| DGP2 - Trend | 0.621 | 0.184 | 0.0878 | 0.0579 |
| DGP2 - Season | 0.838 | 1.032 | 0.6156 | 0.3412 |
| DGP3 - Signal | 11.307 | 3.041 | 0.706 | 0.439 |
| DGP3 - Trend | 10.509 | 0.364 | 0.274 | 0.293 |
| DGP3 - Season | 0.900 | 2.683 | 0.444 | 0.278 |
| DGP4 - Signal | 11.829 | 11.430 | 20.548 | 2.877 |
| DGP4 - Trend | 11.111 | 12.328 | 13.431 | 5.210 |
| DGP4 - Season | 5.364 | 3.045 | 11.358 | 2.562 |
| DGP / Component | EC STR | EC BASTION | Width STR | Width BASTION |
|---|---|---|---|---|
| DGP1 Signal | 0.799 | 0.998 | 4.048 | 5.306 |
| DGP1 Trend | 0.616 | 0.970 | 1.934 | 1.993 |
| DGP1 Season | 0.815 | 0.999 | 2.114 | 3.314 |
| DGP2 Signal | 0.869 | 0.989 | 2.179 | 2.779 |
| DGP2 Trend | 0.716 | 0.976 | 0.992 | 0.854 |
| DGP2 Season | 0.793 | 0.956 | 1.187 | 1.925 |
| DGP3 Signal | 0.940 | 0.995 | 3.017 | 4.589 |
| DGP3 Trend | 0.812 | 0.929 | 1.621 | 1.959 |
| DGP3 Season | 0.847 | 0.981 | 1.396 | 2.630 |
| DGP4 Signal | 0.679 | 0.981 | 5.148 | 5.606 |
| DGP4 Trend | 0.668 | 0.939 | 2.296 | 2.246 |
| DGP4 Season | 0.623 | 0.999 | 2.852 | 3.360 |
元数据与概览(引言与报告概览)
逐节深度解读
图表与图形(深度解读)
估计与贝叶斯实现细节(第4节与附录 B)
风险因素与方法限制(基于报告内容的审慎视角)
结论性综合
若需我可以:
2601.18052.pdf