How kth-nek5000/deepturbulence works
This repository generates time series from the nine-equation Moehlis wall-turbulence model, trains MLP or LSTM predictors, and applies trained models to test series. The README also documents amplitude plotting and flow-field visualization. MATLAB source excerpts were not provided, so the documented generation and visualization roles are shown without inferring internal calls; training and prediction links follow the sampled Python code.
Main entry points: trained_nn_models Trained models · predict_using_mlp.py MLP prediction · predict_using_lstm.py LSTM prediction · moehlis_model_script.m Generate series
Components 14
Data generation
- Generate series
Data generator (Moehlis model)/moehlis_model_script.m - Moehlis ODE
Data generator (Moehlis model)/moehlis_model_odefun.m - Batch generation
Data generator (Moehlis model)/moehlis_data_gen.m
Model training
- Train MLP
Neural networks models/train_mlp_model.py - Train LSTM
Neural networks models/train_lstm_model.py - Trained models
Neural networks models/trained_nn_models
Prediction
- MLP prediction
Neural networks models/predict_using_mlp.py - LSTM prediction
Neural networks models/predict_using_lstm.py
Data inspection
- Plot amplitudes
Data generator (Moehlis model)/plot_amplitudes.m - Visualize flow
Data generator (Moehlis model)/visualize_fields.m
Other components
- Researcher
- Time-series data
- Test series
- Predicted series
Connections 16
- Researcher → Generate series: runs
- Researcher → Batch generation: runs
- Generate series → Time-series data: generates
- Batch generation → Time-series data: gathers series
- Time-series data → Train MLP: loads
- Time-series data → Train LSTM: loads
- Train MLP → Trained models: saves model
- Train LSTM → Trained models: saves model
- Test series → MLP prediction: loads
- Test series → LSTM prediction: loads
- Trained models → MLP prediction: loads model
- Trained models → LSTM prediction: loads model
- MLP prediction → Predicted series: saves predictions
- LSTM prediction → Predicted series: saves predictions
- Researcher → Plot amplitudes: runs
- Researcher → Visualize flow: runs