CNN branch | Quantity/function |
---|---|
Time sequence dimension | 34 |
Input feature dimension | 1096 |
Convolution layer | 5/ReLU |
Convolution kernel size | 7 × 7 |
Convolution kernel Quantity | 128 |
Pooling layer | 5 |
Pooling size | 5 × 5 |
Pooling method | Max-pooling |
Full connection layer units | 128 |
Dropout rate | 0.2 |
GRU branch | Quantity/function |
---|---|
Time dimension(s) | 34 |
Input feature dimension(d) | 1096 |
GRUs | 256 |
Full connection layer units | 256 |
Dropout rate | 0.2 |
Full connection layer | Quantity/function |
---|---|
Layers | 3/ReLU |
Units on the first layer | 128 |
Units on the second layer | 64 |
Units on the last layer | 1 |
Loss function | Improved FL |
Optimizer | Adam |
Initial learning rate | 4 × 10–3 |
The first decay rate (β1) | 0.8 |
The second decay rate (β2) | 0.5 |
Epsilon (ε) | 2 × 10–9 |
Total parameters of the model | 36,284 |