Application of Physics-Informed Machine Learning Techniques for Power Grid Parameter Estimation

Power grid parameter estimation involves the of unknown parameters, such as inertia and damping coefficients, from observed dynamics. In this work, we present physics-informed machine learning algorithms for power system problem. First, propose a novel algorithm to solve based on Sparse Identification Nonlinear Dynamics (SINDy) approach, which uses sparse regression infer parameters that best describe data. We then compare its performance against another benchmark algorithm, namely, neural networks (PINN) approach applied estimation. perform extensive simulations IEEE bus systems examine aforementioned algorithms. Our results show SINDy outperforms PINN in estimating over wide range (including high low systems) architectures. Particularly, case slow dynamics system, proposed struggles accurately determine parameters. Moreover, it is extremely efficient computationally so takes significantly less time than thus making suitable real-time Furthermore, an extension scenario where operator does not have exact knowledge underlying model. also decentralised implementation only requires limited information exchange between neighbouring nodes grid.