Prediction and Optimization of Power Grid Cross Section Exceeding Limits based on Generative Adversarial Networks

Exceeding the cross-sectional power limit of the power grid poses significant risks to the safety and stability of power system operations. Traditional prediction methods based on models and historical data face challenges in accurately capturing the nonlinear characteristics of power grids under complex environments. This study proposes a novel method using Generative Adversarial Networks (GANs) to predict and optimize cross-sectional power violations. GANs are utilized to train a generator network to learn the distribution of power grid operation data and simulate realistic scenarios, combined with a discriminator network in an adversarial framework to enhance prediction accuracy. Additionally, a power optimization scheduling strategy is developed to minimize the probability of exceeding limits using the prediction results. Experimental findings reveal that the GAN model significantly outperforms ARIMA and ANN models in cost optimization, achieving an average reduction of 1919 yuan/month and a maximum of 1984 yuan/month. These results demonstrate the superior performance of GAN models in predicting and optimizing power grid operations.