Deep learning-based contingency analysis in a deregulated power market

This paper presents a novel method for CA in DPM using deep neural networks (DNNs). The proposed work to screen the over voltage's and overpower flows developed the voltage performance index (PIV) and active power performance index (PIMW), which were calculated to indicate contingency severity. The performance indices were predicted using a multilayer perceptron (MLP) DNN trained using a backpropagation algorithm. MLP categorizes contingencies as either critical or non-critical, with essential cases of test ranked based on severity. The proposed MLP DNN was trained using different load scenarios to predict severe contingencies based on performance index values. The proposed technique was evaluated using the IEEE 30-bus system, and load flow studies were conducted using the PANDAPOWER Python-based software. The results show that the trained DNN can predict unforeseen contingencies under various load conditions. The predictions of the proposed method were in close agreement with those of conventional load flow studies, exhibiting a minimal error rate of approximately 2%. This method significantly helps system operators to predict severe contingencies, thereby facilitating congestion control in the DPM