Table 2 Neural network model architecture parameters.
Layer name | Neuron count | Activation function | Dropout rate | Parameter count |
|---|---|---|---|---|
Input layer | Variable | - | - | - |
LSTM layer 1 | 128 (×2 directions) | tanh/sigmoid | 0.2 | ~ 130 K |
LSTM layer 2 | 64 (×2 directions) | tanh/sigmoid | 0.2 | ~ 82 K |
LSTM layer 3 | 32 (×2 directions) | tanh/sigmoid | 0.2 | ~ 25 K |
Attention layer | n (sensor count) | softmax | - | ~ 2 K |
Fully connected layer 1 | 64 | ReLU | 0.3 | ~ 8 K |
Fully connected layer 2 | 32 | ReLU | 0.3 | ~ 2 K |
Output layer | k (prediction steps) | Linear | - | ~ 1 K |
