An ensemble score filter for tracking high-dimensional nonlinear dynamical systems

Code repository for the paper:
Ensemble Score Filter for Tracking High-dimensional Nonlinear Dynamical System
Feng Bao, Zezhong Zhang, Guannan Zhang
Computer Methods in Applied Mechanics and Engineering, 2024, [paper]

Usage

1. data contains the initial state of L96 model and the random shock profile.
   • gen_shock.ipynb generates the random shock profiles.
   • gen_state_init.ipynb generates the initial states for L96 model with different dimensions.
   • Generate the initial state first before running the one-million-dimensional problem.
2. fine_tune_EnSF is the folder for fine-tuning hyperparameters of EnSF (fine_tune_LETKF for LETKF).
   • gen_filter_param.ipynb generates parameter combinations in para_EnSF.csv
   • param_problem.csv is the filtering problem setting for fine-tuning.
   • run_all_EnSF.ipynb runs all combinations of parameters stored in results.
   • results/result_summary.ipynb aggregates all the results to results/final_rmse_EnSF.csv and results/rmse_EnSF.csv
   • to_grid.ipynb generates 2D grid plot for fine-tuning results.
3. run_all_EnSF is the folder for repetitive runs of different filtering settings and parameter combinations (run_all_LETKF for LETKF).
   • param_EnSF.csv and param_problem.csv are user input settings for filtering problem and filter.
   • gen_param_all.ipynb generates all the problem/filter combinations in param_combined.csv with different random seeds for repeated experiments.
   • run_all_EnSF.ipynb runs all problem/filter combinations and stores the results in result
4. run_single_EnSF is the folder for a single run of the filter (run_single_LETKF for LETKF).
   • run_single_EnSF.ipynb runs the problem/filter combination in param_combined.csv and stores the results in result
5. legacy_code is the legacy code for the originally uploaded EnSF.

Citation

If you find the idea or code of this paper useful for your research, please consider citing us:

@article{bao2024ensemble,
  title={An ensemble score filter for tracking high-dimensional nonlinear dynamical systems},
  author={Bao, Feng and Zhang, Zezhong and Zhang, Guannan},
  journal={Computer Methods in Applied Mechanics and Engineering},
  volume={432},
  pages={117447},
  year={2024},
  publisher={Elsevier}
}