Diffusion Model-based Parameter Estimation in Dynamic Power Systems
This work addresses the non-uniqueness challenge in parameter estimation for dynamic power systems, offering a data-driven framework with potential broader scientific applications, though it is incremental as it builds on diffusion models for a specific domain problem.
The paper tackles the ill-posed parameter estimation problem in dynamic power systems by introducing the Joint Conditional Diffusion Model-based Inverse Problem Solver (JCDI), which reduces parameter estimation error by 58.6% compared to a single-condition model and achieves root mean square errors below 4*10^(-3) in replicating system responses.
Parameter estimation, which represents a classical inverse problem, is often ill-posed as different parameter combinations can yield identical outputs. This non-uniqueness poses a critical barrier to accurate and unique identification. This work introduces a novel parameter estimation framework to address such limits: the Joint Conditional Diffusion Model-based Inverse Problem Solver (JCDI). By leveraging the stochasticity of diffusion models, JCDI produces possible solutions revealing underlying distributions. Joint conditioning on multiple observations further narrows the posterior distributions of non-identifiable parameters. For the challenging task in dynamic power systems: composite load model parameterization, JCDI achieves a 58.6% reduction in parameter estimation error compared to the single-condition model. It also accurately replicates system's dynamic responses under various electrical faults, with root mean square errors below 4*10^(-3), outperforming existing deep-reinforcement-learning and supervised learning approaches. Given its data-driven nature, JCDI provides a universal framework for parameter estimation while effectively mitigating the non-uniqueness challenge across scientific domains.