A physics-informed frequency security constrained unit commitment

• PINN-based model predicts frequency nadir for unit commitment. • Physics-informed constraints ensure MILP compatibility and interpretability. • Big-M linearization encodes ReLU for exact integration into MILP-based UC framework. • PINN method outperforms ML/DL in cost efficiency and frequency security. • BESS integration enhances frequency stability and reduces UC operating cost. With the significant growth in renewable energy integration, system inertia has substantially decreased, necessitating the incorporation of frequency security constraints into the conventional unit commitment (UC) problem. Since the system frequency response is a highly nonlinear and dynamic process, this study proposes a frequency nadir prediction method based on a physics-informed neural network (PINN) and is constrained in the UC problem. The linearization technique is then applied to integrate the frequency security constraint. By embedding the system dynamics equations within the frequency constraint, this study develops a physics- and MILP-based frequency-aware UC model. Test results demonstrate that the proposed method effectively achieves the lowest operating cost while strictly complying with the frequency nadir constraint. Furthermore, the inclusion of battery energy storage systems (BESSs) in the UC is further optimized for better economic performance. This model successfully bridges the gap between analytical modeling and data-driven prediction for frequency dynamic constraints, providing a reliable and cost-effective operational solution for power systems with high penetration of renewable energy resources.