Rolling Optimization for VPP in Spot or Ancillary Service Markets Based on Deep Learning Forecasts

This paper presents a comprehensive framework for Virtual Power Plant (VPP) participation in spot and ancillary service markets, integrating deep learning-based forecasting with rolling optimization strategies. The proposed approach combines Transformer-GRU hybrid architecture for renewable energy forecasting and LSTM networks for electricity price prediction, achieving forecast accuracies exceeding 96%. A three-stage optimization framework is developed: day-ahead forecast declaration, intra-day rolling optimization, and intra-day dispatch order optimization. The rolling optimization strategy maximizes VPP profitability by coordinating various distributed energy resources, including renewable generation, energy storage systems, and flexible loads. Case studies demonstrate that the proposed approach improves VPP profitability by 16.25% compared to non-optimized operations, while maintaining rapid response capabilities to grid dispatch orders within 0.1 seconds. The results validate the effectiveness of the proposed framework in enhancing both economic performance and operational efficiency of VPPs in modern electricity markets.