Prepared datasets and models for modeling orientational variation in heat plume prediction in groundwater.
Models were trained with 1HP NN equivariance.
File name explanation:
4d
The (used) dataset encompasses only cardinal flow directions.
rd
The (used) dataset encompasses random flow directions in the 2D plane.
1000dp
The (used) dataset consists of 1000 data points.
100dp
The (used) dataset consists of 100 data points.
pksi
The input data fields: liquid pressure, permeability, position of the heat pump, and normalized distance to the heat pump.
BaselineCNN
The model is trained without any modification to architecture/training procedure.
DataAugmentation
The model is trained using data augmentation where rotated variations of the original data were added to the training data.
OrientedBoxes
For this model, during both training and inference, the input data is aligned to a chosen orientation.
ECNN
The model uses the Equivariant Convolutional Neural Network (ECNN) architecture.
For usage instructions see 1HP NN equivariance.
