Research on Online Monitoring of AGC Frequency Regulation Capacity in Regional New Energy Power Systems

With the rapid increase in the proportion of renewable energy in regional power systems, frequency regulation faces growing challenges, and traditional static estimation methods fail to accurately capture the dynamic characteristics of AGC (Automatic Generation Control) frequency regulation capacity. To address the impact of renewable energy output fluctuations and forecasting errors, this paper proposes a hybrid estimation method based on a "data-state-physical constraint" three-layer modeling framework. The method employs Long Short-Term Memory (LSTM) networks to dynamically predict the frequency response capability of generating units, integrating physical constraints to achieve accurate estimation of AGC regulation capacity. To enhance real-time performance and robustness, a sliding time window mechanism and Bayesian online parameter update strategy are designed, along with anomaly detection and fallback mechanisms. Validation based on real dispatch data and simulation platforms demonstrates that the proposed method outperforms traditional static and purely data-driven models in estimation accuracy, dynamic response capability, and stability. Under typical scenarios of renewable energy fluctuations, the model can swiftly and accurately track capacity changes, supporting dispatch decisions and ancillary service optimization, and exhibits strong engineering application potential.