Symbolic Regression for Data-Driven Dynamic Model Refinement in Power Systems

This paper describes a data-driven symbolic regression identification method tailored to power systems and demonstrated on different synchronous generator (SG) models. In this work, we extend the sparse identification of nonlinear dynamics (SINDy) modeling procedure to include the effects of exogenous signals (measurements), nonlinear trigonometric terms in the library of elements, equality, and boundary constraints of expected solution. We show that the resulting framework requires fairly little in terms of data, and is computationally efficient and robust to noise, making it a viable candidate for online identification in response to rapid system changes. The SINDy-based model identification is integrated with the manifold boundary approximation method (MBAM) for the reduction of the differential-algebraic equations (DAE)-based SG dynamic models (decrease in the number of states and parameters). The proposed procedure is illustrated on an SG example in a real-world 441-bus and 67-machine benchmark.

Automated summary

This paper presents a data-driven symbolic regression identification method designed for power systems, which extends the SINDy modeling procedure to include exogenous signals and nonlinear trigonometric terms. The resulting framework is shown to require minimal data, be computationally efficient, and robust to noise, making it a feasible option for online identification in response to rapid system changes. The proposed method is illustrated on a real-world benchmark example, demonstrating its effectiveness in reducing the differential-algebraic equations-based SG dynamic models.