Research on Fast Calculation Method for Relay Protection Setting Mismatch Points of Lines Based on Deep Learning

In complex ring networks, a significant amount of time is typically required to determine the coordination relationships of backup protection settings during online adjustment, which affects the efficiency of the online setting process. To address this, the paper introduces a data-driven approach by applying artificial intelligence technology—specifically, for the first time—into the field of online setting. A fast computation scheme for online backup protection setting based on Generative Adversarial Networks (GANs) is proposed. Firstly, a conditional GAN based on the Wasserstein distance is constructed, using system operating modes as conditional labels. The coordination pairs of backup protection settings are transformed into matrix indices, forming a backup protection coordination matrix, which serves as real sample data for GAN training. The generator and discriminator networks of the GAN are primarily composed of Convolutional Neural Networks (CNNs), and batch normalization is applied to the outputs of the neural networks to improve training stability. Next, the specific implementation of the GAN-based online setting scheme for backup protection is presented. Parallel computing techniques are employed to accelerate the computation of both the CNN and the setting values, further enhancing the efficiency of the online adjustment process. Finally, the IEEE 39-bus system is used as an example to compute backup protection coordination matrices under various operating conditions, thus constructing the dataset required for GAN training and validating the proposed scheme. Simulation results show that the proposed method can learn the complex coordination relationships among various backup protection settings from real sample data via GANs. It can generate the corresponding coordination matrix according to different operating modes, thereby achieving fast computation for online backup protection setting.