Table 1 WMF-Traffic Component Configuration.
From: Multi-scale Wavelet-Mamba framework for spatiotemporal traffic forecasting
Component | Parameter | Value |
|---|---|---|
MWD | Decomposition levels (L) | 3 |
Wavelet basis | Daubechies-4 | |
Energy threshold | 0.01 | |
Padding mode | symmetric | |
WTC | Base kernel size (\(k_0\)) | 7 |
Maximum kernel size (\(k_{\max }\)) | 15 | |
Scaling factor (\(\rho\)) | 1.5 | |
Attention dimension (\(d_\alpha\)) | 16 | |
Gating activation | Sigmoid | |
T-Mamba | Hidden dimension (\(d_h\)) | 64 |
Feature dimension (\(d_i\),\(d_o\)) | 32 | |
Context dimension (\(d_c\)) | 8 | |
Temporal scales (K) | 4 | |
Mixing coefficient (\(\gamma\)) | 0.1 | |
Selection smoothing | Softplus | |
FPA | Period priors\(\{p\}\) | {24, 168, 8760} hours |
Energy scaling (\(\beta\)) | 0.5 | |
Frequency attention dim | 32 | |
Phase preservation (\(\lambda _\pi\)) | 0.3 | |
Training | Batch size | 32 |
Learning rate | 0.001 | |
Weight decay | 0.0001 | |
Dropout rate | 0.1 | |
Training epochs | 1000 |
