A DRL-Based Approach for System Frequency Response Model Calibration

The system frequency response model (SFR) has been widely adopted in power grid dynamics analysis for describing the average frequency response of all elements in a power network. It is a challenging problem to accurately calibrate the parameters of the SFR model for a real bulk power grid. This paper proposes a deep reinforcement learning (DRL) based approach for adaptive SFR model parameter calibration. The frequency responses of bulk power systems under multiple contingencies are considered. Detailed parameter selection, architecture design, key component definition, and environment setup associated with the DRL implementation are analyzed and illustrated. The proposed DRL method for SFR parameter calibration is verified using a real bulk power system with thousands of buses in China considering the faults of ultra-high voltage direct current (UHVDC) lines. It is shown that the proposed method can optimize the SFR model parameters efficiently and accurately.