Smart Grid Stability Prediction Based on Deep Learning Models on the Spark Platform

As the electricity demand continues to grow, smart grids generate vast amounts of data, posing a significant challenge for effectively mining this data to monitor stability and predict outcomes. Given the difficulty of processing large-scale power grid data in standalone systems and the inherent limitations of model accuracy and robustness, this paper proposes a power grid stability prediction system based on the Spark cloud computing platform and deep learning models. Specifically, the paper constructs nine power grid stability models, incorporating traditional models such as logistic regression and XGBoost, alongside deep learning models like ResNet and DenseNet. It also employs ensemble learning strategies, including Voting, Boosting, and Stacking methods, to integrate and apply these models. The results demonstrate that the deep learning ensemble model outperforms any individual model, with the stacking model achieving an accuracy of 99.52%. Additionally, the paper confirms that data processing efficiency on the Spark platform surpasses that of standalone systems, particularly with large datasets. This study holds significant value for smart grid stability prediction, aiding power companies in risk prevention. Future work will focus on optimizing model structures and integrated learning strategies to enhance Spark platform performance and further improve prediction accuracy.