Table 2 Model details and descriptions
From: Data-driven dynamic modeling for inverter-based resources using neural networks
Model | Number of parameters | Description |
|---|---|---|
LSTM-16 | 1922 | The LSTM with 16 hidden layers. |
RNN | 1946 | The RNN with 24 hidden layers. |
GRU | 1964 | The GRU with 18 hidden layers. |
TCN | 1946 | TCN with 3 convolutional layers (13 channels each). |
MLP | 1924 | MLP with 7 layers (17 neurons each), using Hardtanh as activation function. |
PINN | 1924 | A PINN enforcing the physics of the second-generation generic model as hard constraints, based on the architecture in ref. 51. |
Transformer | 1939 | The Transformer with 5 encoder and decoder layers, 7 multi-head attention heads, an input feature dimension of 7, an 8-dimensional feed-forward network, and a 0.1 dropout rate. |
LSTM-8 | 578 | The LSTM with eight hidden layers. |
LSTM+Inv. | 588 | LSTM-8 with an inverter model. |
LSTM+Cro. | 1207 | LSTM-8 with Cross-layer. |
RNN+Cro.+Inv. | 1628 | RNN (8 hidden layers) with cross-layer and inverter model. |
LSTM+DCN+Inv. | 652 | LSTM-8 with DCN and inverter model. |
LSTMCI | 1940 | LSTM-8 with Cross layer and inverter (can also be marked as LSTM+Cro.+Inv.). |
