Imbalance Learning-Based Stability Prediction for Virtual Power Plant-Enabled Smart Grids

The widespread adoption of virtual power plant (VPP) technology has driven significant advancements in smart grids by enabling real-time coordination and enhanced energy management of distributed resources. However, the integration of numerous distributed energy resources and dynamic loads introduces substantial complexity, particularly in maintaining grid stability. One of the main challenges is the class imbalance inherent in smart grid data, where stable states vastly outnumber unstable ones, making it difficult for traditional learning models to perform effectively. To address this issue, this paper proposes a deep learning-based stability prediction method specifically designed to target the class imbalance. The approach incorporates z-score normalization to mitigate disparities in the scale of input data and uses a focal loss function to enhance the model's sensitivity to minority (unstable) samples. A deep neural network (DNN) with multiple hidden layers is trained to classify grid states as either stable or unstable. Simulation results show that the proposed method consistently outperforms classical machine learning models and other deep learning techniques, confirming its robustness and effectiveness in predicting smart grid stability.