A unified wind power prediction framework combined with individual wind turbine operation status and error correction

Higher precision wind power prediction is needed as a reliable basis for grid scheduling. To achieve synchronous prediction of the power of all wind turbines in the entire wind farm, this paper proposes a prediction strategy that combines unified prediction and individual wind turbine error correction . The proposed unified prediction framework includes a spatiotemporal joint prediction model (STJPM) and multiple individual wind turbine error correction (EC) models. Proposed STJPM consists of temporal convolutional network (TCN) and 2D convolutional gated recurrent unit (ConvGRU), which can share the loss function for end-to-end network training. Initially, the numerical weather prediction (NWP) grid and gradient plus inverse distance squared (GIDS) were used to obtain predicted meteorological elements for each wind turbine location, and the converted power sequence was obtained based on the wind speed-power conversion curve (WPCC) of the wind turbine. Subsequently, the historical power and converted power sequences from the wind turbine monitoring system enter STJPM in chronological order, and perform joint learning of the spatiotemporal characteristics of all wind turbine power sequences to obtain preliminary wind power prediction sequences. Finally, a combined GRU (Com-GRU) model is proposed as a individual wind turbine EC model, further improving the prediction accuracy of all wind turbines . The prediction strategy proposed in this article achieves spatiotemporal joint training of irregular wind turbine arrangements in the form of a virtual wind farm. Compared to simply improving the prediction model (increase by 1–3 %), the error correction strategy based on the EC model has achieved a more significant performance improvement (increase by more than 5 %).

Hali tarjima qilinmagan