基于双通道网络和迁移学习的超短期风电功率预测OA

An ultra short term wind power prediction method based on dual channel network and transfer learning is proposed to address the problem of insufficient historical data for newly built wind farms,which makes it difficult to achieve high-precision wind power prediction.In this method,the features are first extracted from the meteorological data and power data through the dual-channel network.Among them,the meteorological channel adopts the attention mechanism to dynamically assign weights to meteorological factors.After that,the convolutional neural network is used to extract high-dimensional meteorological features in the data.The power channel uses a gated recurrent unit to extract the change trend of features and predict future power generation.Then,the features extracted by the dual-channel are then combined and regressed to the wind power via a feedforward neural network.Finally,the training strategy of transfer learning is used to pre-train the model on the wind farm data with sufficient data.After pre-training,transfer the model to a target wind farm with insufficient data volume.Simulation results show that the proposed method outperforms other predictive models.Moreover,it can effectively improve the problem of poor prediction accuracy caused by insufficient data volume of new wind farms.