Active integrated protection method for substation DC system based on sequential network and deep learning

In response to the problems of outdated setting of fault level classification, poor time series analysis ability, and high computational complexity in the current active comprehensive protection method for substation DC system, which leads to high rejection rate of protection equipment and low accuracy of fault detection and recognition, a proactive comprehensive protection method for substation DC system based on time sequence network combined with deep learning is proposed. Construct a mathematical model of time series information and an active time series model of the DC system, complete fault time series analysis and protection device action time series analysis. Select the Recurrent Neural Network (RNN) to complete deep learning and identification of faults in the substation DC system. Applying data analysis results, the current change rate is used as the starting element to distinguish the fault status of equipment from load fluctuations, determine the fault level, and start the protection device. Build a simulation model and complete argumentation analysis. The results show that this method can further improve the accuracy of fault category identification to more than $\mathbf{9 5 \%}$, and reduce the rejection rate of the protection device to less than $\mathbf{0. 7 3 \%}$.