Leveraging Deep Learning to Improve Performance of Distributed Optimal Frequency Control Under Communication Failures

This paper proposes a deep learning approach to overcome the impacts of communication failures on the performance and convergence rate of the distributed optimal frequency control (DOFC) for power systems. Novel features of the proposed framework are fourfold. First, the nonlinear model of the DOFC is developed to consider for nonlinearities of power flows. Second, the long short-term memory (LSTM) algorithm is used for dynamic model estimation during communication failures. Next, the LSTM-based DOFC method is introduced to cope with the impact of communication failures on the performance of the distributed control strategy. Finally, we prove the convergence of LSTM-DOFC and show that the algorithm has superior performance compared to the linearized prediction methods, such as autoregressive-moving-average models. Simulations on two real-world power systems are carried out to demonstrate the effectiveness of the proposed framework.