A Deep Learning Approach for Inertia Estimation in Renewable Integrated Systems: Combining LRCN with PINN for Real Time Application

Frequency stability and grid reliability are faced with novel challenges as a result of the substantial reduction in system inertia produced by the increasing incorporation of renewable energy sources(RES), such as solar photovoltaics (PV) and wind power, in modern power systems. Since renewable energy sources dominate modern power systems, estimating inertia using mathematical models becomes more difficult. This study proposes two combined machine learning strategies to estimate inertia in hybrid networks. The traditional Weighted Least Squares (WLS) method and the Long-Recurrent Convolutional Neural Network (LRCN) in addition with Physics Informed Neural Networks (PINN) are incorporated here. Under various operational conditions, case studies are carried out using the IEEE 30 bus system using PSS/E with hybrid networks that include conventional generators, wind power, and solar PV. The precision of the conventional WLS approach, LRCN and combined LRCN and PINN are 92 to 95%, 95 to 98%, and 98 to 99.5% respectively. The result shows that the combined approach based on LRCN and PINN outperforms WLS and LRCN, achieving higher accuracy, robustness to measurement noise and adaptability to complex grid configurations. These results demonstrate the drawbacks of conventional methods and the potential of specialized machine learning models, such as LRCN and PINN, to address emerging problems in renewable-integrated power systems.