Interactive Power Prediction Model for Virtual Power Plant Based on Wave-Attention Network

Virtual power plant technology has emerged as a crucial technical solution for addressing the challenges related to scheduling of new energy and distributed power grids. However, the interactive power characteristic of virtual power plant is affected by the power generator, renewable energy and load, which results in the characteristics of non-linearity, time series coupling and time variation. To better depict the interactive power characteristics of virtual power plants, this paper proposes a wave-attention network based interactive power prediction model. The proposed wave-attention network adopts casual convolutional layers and residual connections to extract the time series feature of interactive power, and uses attention mechanism to capture the internal relationship of virtual power plant. This prediction model has the advantages of model-free and quick response compared with the traditional optimization methods. Finally, the effectiveness of the proposed method is verified by the analysis of several virtual power plants. The case study shows that the proposed wave-attention network has the advantages of higher prediction accuracy and better robustness compared with other deep learning methods.