Real-time congestion control using cascaded LSTM deep neural networks for deregulated power markets

In deregulated power markets (DPMs), transmission-line congestion has become more severe and frequent than in traditional power systems. This congestion hinders electricity markets from operating in normal competitive equilibrium. The independent system operator (ISO) is responsible for implementing appropriate measures to mitigate congestion and ensure the proper functioning of the power market. This study utilized generation rescheduling (GR) to address congestion in spot or day-ahead power markets. Rapid alleviation of congestion is essential to prevent tripping of overloaded lines. Owing to their computational inefficiency, evolutionary algorithms (EAs) are ineffective for real-time congestion management, necessitating hybrid models to deliver rapid solutions. This paper proposes a hybrid deep neural network (DNN)-based congestion management (CM) approach for real-time congestion control and reduced computation time. The proposed CM system consists of three cascaded long short-term memory (LSTM) DNNs that operate sequentially. These LSTM modules predict congestion status, violated power, and adjusted active power for rescheduling generation. The LSTM-DNN was trained using data from a grey wolf optimization (GWO). This method provides a rapid solution with approximately 98% accuracy for managing congestion in the spot power market. The suggested approach was evaluated on an IEEE 30 bus system and demonstrated to be highly effective.