Hybrid-Driven State Estimation With Adaptive Cross-Coupled Priors: Enhancing Data Representation and Model...

Area of Science:

• Control Systems Engineering
• Machine Learning
• Signal Processing

Background:

• State estimation is crucial for understanding system dynamics.
• Integrating model-driven and data-driven methods offers potential for improved robustness.
• Limited data and model uncertainties pose significant challenges in hybrid estimation.

Purpose of the Study:

• To propose an unsupervised hybrid estimation framework (AMD) that robustly integrates model-driven and data-driven approaches.
• To enhance state estimation accuracy under conditions of limited data and model uncertainties.
• To develop a framework adaptable to complex nonlinear systems.

Main Methods:

• Developed an adaptive model-driven and data-driven (AMD) framework using Bayesian inference.
• Implemented an adaptive cross-coupled prior mechanism for integrating prior information.
• Introduced a two-stage fusion strategy: initial hard fusion followed by adaptive soft fusion.
• Incorporated a dynamic bilinear recurrent module for nonlinear transition dynamics.
• Utilized a nonidentical training-testing strategy and an unsupervised hybrid learning objective.

Main Results:

• AMD demonstrated competitive or superior estimation accuracy compared to state-of-the-art methods.
• The framework showed high performance in underdetermined estimation, model mismatch, and dynamic disturbances.
• AMD effectively leveraged limited information through complementary fusion.
• Enhanced robustness to imperfect model priors was achieved via adaptive soft fusion.

Conclusions:

• The proposed AMD framework offers a robust solution for challenging state estimation problems.
• AMD's adaptability enhances both data representation and model robustness.
• This approach effectively utilizes complementary information for improved state estimation.