RCUKF is a hybrid framework that combines Reservoir Computing (RC) and the Unscented Kalman Filter (UKF) for robust, data-driven state estimation in nonlinear dynamical systems. This approach avoids explicit modeling of system dynamics and is especially powerful for chaotic and partially observed systems.

• ✅ Model-free learning using Reservoir Computing (Echo State Network)
• ✅ Filtering with Unscented Kalman Filter
• ✅ No backpropagation: fast training using ridge regression
• ✅ Supports chaotic systems like Lorenz, Rössler, Mackey-Glass
• ✅ Plug-and-play modular design

RCUKF/
├── benchmarks/ # RC vs RCUKF RMSE comparison and plots
├── data_gen/ # Noisy synthetic data generators
├── demos/ # End-to-end implementations of RCUKF
├── rcukfpy/ # Core implementation (RC, UKF, RCUKF)
├── utils/ # RMSE calculator and helper utils
├── LICENSE
└── README.md

git clone https://github.com/kmranrg/rcukf.git
cd rcukf

pip install numpy matplotlib

Compare RC-only vs RC+UKF on the Lorenz system:

python -m benchmarks.RC_vs_RCplusUKF_Lorenz

Other systems:

python -m benchmarks.RC_vs_RCplusUKF_Rossler
python -m benchmarks.RC_vs_RCplusUKF_MackeyGlass

python demos.RCplusUKF_LorenzImplementation
python demos.RCplusUKF_MackeyGlassImplementation
python demos.RCplusUKF_RosslerImplementation

• ReservoirComputer: Trains a readout model to approximate system dynamics.
• UnscentedKalmanFilter: Uses sigma points to filter noisy observations.
• RC_UKF: Wraps both modules to predict system state using data + measurements.

Pull requests are welcome! Please open an issue first to discuss any major changes or improvements.

Developed with ❤️ by Kumar Anurag Website | LinkedIn