Decentralized Bayesian Inference for Inertia Estimation in Modern Power System Using Ambient Measurements

Tracking inertia in real-time has emerged as a crucial topic for the modern low-inertia power system. However, in practice, large disturbances are infrequent, posing a challenge for routine monitoring of inertia through fault-based methods. To tackle this issue, we propose a Bayesian method for tracking inertia in real-time for both synchronous and asynchronous generators, utilizing only ambient measurements. To circumvent the curse of dimensionality inherent in Bayesian inference, an ambient measurement-based decentralized Bayesian framework is devised. When relying on ambient measurements, the inertia estimation generally requires data from a much longer time window than methods that depend on large disturbances. This extended time window, however, introduces severe error accumulations that compromise the accuracy of the decentralized model. To address this issue, we further incorporate the variational mode decomposition (VMD) to recover the precision of the decentralized scheme. Finally, we propose a cost-effective multiple-importance-sampling (MIS) algorithm to obtain the non- Gaussian posterior distribution of inertia. Simulation results demonstrate the excellent performance of the proposed method. It can not only precisely estimate inertia under continuous load changes but also maintain high accuracy even in the presence of discrete load fluctuations, including load jumps.