Configuration Parameters

Detailed explanation of the parameters used in the YAML configuration file. The configuration is structured into different sections for ease of use.

---

Data

• networks: (List of Strings) Specifies the network topologies to use during training

  • Example: ["case300_ieee", "case30_ieee"]

• scenarios: (List of Integers) Defines the number of scenarios to use for each network specified.

• Example: [8500, 4000]

• normalization: (String) Normalization method for data.

  • Options:
    • minmax: Scales data between the minimum and maximum values.
    • standard: Standardizes data to have zero mean and unit variance.
    • baseMVAnorm: Divides data by a baseMVA value, which is the maximum active/reactive power across the network.
    • identity: Leaves data unchanged.
  • Example: "baseMVAnorm"

• baseMVA: (Integer) The base MVA value specified in the original matpower casefile, needed for baseMVAnorm normalization.

  • Default: 100

• mask_type: (String) Masking strategy.

  • Options:
    • rnd: Mask each feature with a certain probability that needs to be specified (mask_ratio).
    • pf: Power flow problem setup.
    • opf: Optimal power flow problem setup.
    • none: No masking.
  • Example: "rnd"

• mask_value: (Float) Value used to mask data during training.

  • Default: 0.0

• mask_ratio: (Float) Propability of each feature to be masked, needs to be specified only when mask_type is rnd.

  • Default: 0.5

• mask_dim: (Integer) Number of features to mask.

  • Default: 6 (Pd, Qd, Pg, Qg, Vm, Va)

• learn_mask: (Boolean) Specifies whether the mask value is learnable.

  • Default: False

• val_ratio: (Float) Fraction of data used for validation.

  • Default: 0.1

• test_ratio: (Float) Fraction of data used for testing.

  • Default: 0.1

Model

• type: (String) Specifies the type of model architecture.

  • Example: "GPSconv"

• input_dim: (Integer) Input dimensionality of the model.

  • Default: 9 (Pd, Qd, Pg, Qg, Vm, Va, PQ, PV, REF)

• output_dim: (Integer) Output dimensionality of the model.

  • Default: 6 (Pd, Qd, Pg, Qg, Vm, Va)

• edge_dim: (Integer) Dimensionality of edge features.

  • Default: 2 (G, B)

• pe_dim: (Integer) Dimensionality of positional encoding.

  • Example: 20 (Length of random walk)

• num_layers: (Integer) Number of layers in the model.

  • Example: 6

• hidden_size: (Integer) Size of hidden layers.

  • Example: 256

• attention_head: (Integer) Number of attention heads in the model.

  • Example: 8

• dropout: (Float) Model dropout probability

  • Default: 0.0

Training

• batch_size: (Integer) Number of samples per training batch.

  • Example: 16

• epochs: (Integer) Number of training epochs.

  • Example: 100

• losses: (List of Strings) Specifies the loss functions to use during training.

  • Available options:
    • MSE: Mean Squared Error.
    • MaskedMSE: Masked Mean Squared Error.
    • SCE: Scaled Cosine Error.
    • PBE: Power Balance Equation loss.
  • Example: ["MaskedMSE", "PBE"]

• loss_weights: (List of Floats) Specifies the relative weights for each loss function.

  • Example: [0.01, 0.99]

Optimizer

• learning_rate: (Float) Learning rate for the optimizer.

  • Example: 0.0001

• beta1: (Float) Beta1 parameter for the Adam optimizer.

  • Default: 0.9

• beta2: (Float) Beta2 parameter for the Adam optimizer.

  • Default: 0.999

• lr_decay: (Float) Learning rate decay factor. `

• lr_patience: (Integer) Number of epochs to wait before applying learning rate decay.

  • Example: 3

Callbacks

• patience: (Integer) Number of epochs to wait before early stopping. A value of -1 disables early stopping.

  • Default: -1

• tol: (Float) Tolerance for validation loss comparison during early stopping.

  • Default: 0

Verbose

• verbose: (Boolean) Provides detailed analysis after training.
  • Default: False

---