Abstract:In this paper, we present a novel approach to state estimation in dynamic systems by combining Physics-Informed Neural Networks (PINNs) with an adaptive Unscented Kalman Filter (UKF). Recognizing the limitations of traditional state estimation methods, we refine the PINN architecture with hybrid loss functions and Monte Carlo Dropout for enhanced uncertainty estimation. The Unscented Kalman Filter is augmented with an adaptive noise covariance mechanism and incorporates model parameters into the state vector to improve adaptability. We further validate this hybrid framework by integrating the enhanced PINN with the UKF for a seamless state prediction pipeline, demonstrating significant improvements in accuracy and robustness. Our experimental results show a marked enhancement in state estimation fidelity for both position and velocity tracking, supported by uncertainty quantification via Bayesian inference and Monte Carlo Dropout. We further extend the simulation and present evaluations on a double pendulum system and state estimation on a quadcopter drone. This comprehensive solution is poised to advance the state-of-the-art in dynamic system estimation, providing unparalleled performance across control theory, machine learning, and numerical optimization domains.

What problem does this paper attempt to address?

The paper aims to address the problem of dynamic system state estimation, particularly the limitations of traditional state estimation methods (such as Kalman filtering) when dealing with highly nonlinear systems. The paper proposes a new hybrid framework that combines Physics-Informed Neural Networks (PINNs) with an Adaptive Unscented Kalman Filter (UKF) to improve the accuracy, robustness, and adaptability of state estimation. ### Overview of the Problem Addressed by the Paper 1. **Integration of PINNs and UKF**: - Utilize the powerful expressive capability of PINNs to capture the physical characteristics of the system and handle uncertainties through the Adaptive Unscented Kalman Filter. - Enhance the uncertainty estimation capability of PINNs through a hybrid loss function and Monte Carlo Dropout. - Introduce a dynamic noise covariance estimation mechanism in the UKF and incorporate model parameters into the state vector to improve adaptability. 2. **Experimental Validation**: - Validate the effectiveness of the hybrid framework through position and velocity tracking experiments, demonstrating significantly improved state estimation accuracy. - Further extend the simulated scenarios, including state estimation for a double pendulum system and a quadrotor drone. 3. **Technical Innovations**: - Propose an improved PINN architecture that combines data-driven learning with physical laws to enhance estimation accuracy. - Introduce an adaptive UKF capable of dynamically adjusting process noise covariance, enhancing the filter's adaptability. - Use Monte Carlo Dropout to quantify model uncertainty, improving the reliability of model predictions. 4. **Performance Comparison**: - Compare with traditional Extended Kalman Filter (EKF), showing that the hybrid framework has higher accuracy, adaptability, and robustness when dealing with nonlinear systems. - Further improve model performance through the enhanced PINN architecture and training process. 5. **Computational Complexity Analysis**: - Evaluate the computational complexity of the hybrid framework, pointing out that although the complexity is higher than traditional Kalman filtering methods, this increase trades off for better uncertainty and nonlinear handling capabilities. - Analyze the computational complexity of the PINN model and UKF, providing specific mathematical formulas and comparisons with traditional Kalman filtering. In summary, the paper proposes a comprehensive and innovative solution by combining the powerful representation capabilities of PINNs with the rigorous handling of uncertainties by UKF, aiming to address key challenges in dynamic system state estimation.

Hybrid State Estimation: Integrating Physics-Informed Neural Networks with Adaptive UKF for Dynamic Systems

Fine-Tuning Hybrid Physics-Informed Neural Networks for Vehicle Dynamics Model Estimation

Vehicle State Estimation Combining Physics-Informed Neural Network and Unscented Kalman Filtering on Manifolds

Physics-aware tuning of the unscented Kalman filter: statistical framework for solving inverse problems involving nonlinear dynamical systems and missing data

Response Estimation and System Identification of Dynamical Systems via Physics-Informed Neural Networks

Accurate and Robust State Estimation via Fusion of Visual-Inertial-UWB with Time Synchronization

Estimating States for Nonlinear Systems Using the Particle Filter

DIDO: Deep Inertial Quadrotor Dynamical Odometry

Physics-Informed Neural Networks for Accelerating Power System State Estimation

Robust Parameter Estimation for Hybrid Dynamical Systems

Efficient Bayesian Physics Informed Neural Networks for inverse problems via Ensemble Kalman Inversion

Estimating States and Model Uncertainties Jointly by a Sparsity Promoting UKF

Are hybrid approaches combining EKF-UKF and artificial neural networks key to unlocking the full potential of sustainable energy technologies and reducing environmental footprint?

Combined State and Parameter Estimation in Level-Set Methods

Physics Informed Neural Network Framework for Unsteady Discretized Reduced Order System

TLIO: Tight Learned Inertial Odometry

Particle Filter Based On Partial Accurate Observation For High Dimensional Stochastic Dynamic System

Legged Robot State Estimation With Invariant Extended Kalman Filter Using Neural Measurement Network

State Parameter Estimation of Intelligent Vehicles Based on an Adaptive Unscented Kalman Filter

Implementation of unknown parameter estimation procedure for hybrid and discrete non‐linear systems

HyResPINNs: Adaptive Hybrid Residual Networks for Learning Optimal Combinations of Neural and RBF Components for Physics-Informed Modeling