Rolling Optimization Method for Day-Ahead and Intra-Day Coordination of Virtual Power Plants Based on Dynamic Weights of Combined Neural Network Predictions

As the integration of large amounts of renewable energy into virtual power plants increases, these energy sources' inherent randomness and volatility present significant challenges for optimization and scheduling. To mitigate the economic losses caused by the uncertainty and variability of renewable energy in VPP (Virtual Power Plant) operations, this paper introduces an optimization approach that combines weight updating with a hybrid neural network model for wind and solar power forecasting. The method aims to improve the accuracy of solar and wind power predictions using a combined neural network, enhanced with the grey wolf optimization algorithm to speed up model training. Furthermore, a weight updating mechanism, along with a coordinated day-ahead and intra-day scheduling strategy, is employed to enhance the economic efficiency of the optimization process. The proposed forecasting and optimization approach is effective and superior in a small-scale virtual power plant system.