Power System Inertia Estimation Based on Long Short Term Memory Network

With the increasingly high penetration of renewable energy into the power system, traditional synchronous generators are gradually substituted by power-electronic-converter-based sources that commonly lack inertia support capability. Since deficient system inertia could cause larger frequency deviation under the same disturbance, commanding the inertia level of the system is beneficial and necessary for the operators to take precautions and ensure stable power supply. This paper puts forward a long short-term memory (LSTM) based inertia estimation method. Through massive training, the LSTM network successfully constructs the relationship between the system inertia and the system frequency sequence as well as the unbalanced power. The hyperparameters of the proposed deep learning model can be determined by the classic grid search. Case studies on the IEEE-39 system verify that the LSTM network can well handle the time-varying sequence and the proposed method can accurately estimate the system inertia.