Diffusion model-based parameter estimation in dynamic power systems.

Area of Science:

• Computational Science
• Electrical Engineering
• Applied Mathematics

Background:

• Parameter estimation is a classical inverse problem often ill-posed due to non-unique solutions.
• This non-uniqueness hinders accurate and reliable identification of system parameters.

Purpose of the Study:

• To introduce a novel parameter estimation framework to overcome the limits of non-unique solutions.
• To leverage diffusion models for capturing parameter distributions conditioned on observations.
• To improve parameter estimation accuracy and efficiency in complex systems.

Main Methods:

• Developed a Joint Conditional Diffusion Model-based Inverse Problem Solver.
• Utilized the stochasticity of diffusion models to generate candidate solutions.
• Employed joint conditioning on multiple observations to refine parameter distributions.

Main Results:

• Achieved a 58.6% reduction in parameter estimation error for composite load models compared to single-condition methods.
• Accurately replicated dynamic power system responses under electrical faults with root mean square errors < 4 × 10^-3.
• Demonstrated comprehensive advantages in calibration and efficiency over existing methods.

Conclusions:

• The proposed framework effectively mitigates the non-uniqueness problem in parameter estimation.
• The data-driven approach offers a general solution applicable across scientific domains.
• The method shows significant improvements in accuracy and efficiency for dynamic power systems.