Power prediction considering NWP wind speed error tolerability: A strategy to improve the accuracy of short-term wind power prediction under wind speed offset scenarios

Short-term wind power prediction is of great significance for wind power participation in day-ahead scheduling. However, unavoidable numerical weather prediction (NWP) errors bring severe challenges to high-precision prediction of wind power, especially in the power peaks and valleys periods, the extreme error is significant. In this regard, this paper proposes a strategy to improve the accuracy of short-term wind power prediction taking into account the wind speed offset scenario and the weighted improved offset loss function (WIOLF). Introducing a multi-level directed acyclic graph structure for identification of wind speed offset scenarios, and a Wasserstein GAN (WGAN-GP) network with gradient penalty is used to solve the problem of sample imbalance. In the power prediction part, WIOLF is integrated into the combination model of temporal convolution network (TCN) combined with multi-head self-attention mechanism (MHSA) to improve its decision-making mechanism, so as to train a wind power offset prediction model to improve the power prediction accuracy in wind speed offset scenarios. The proposed method is applied to several wind farms in Western Inner Mongolia, China, the results show that compared with the direct prediction method, the RMSE and MAE of the proposed method are reduced by 7.41 % and 6.10 %, and the R2 is increased by 9.06 %, respectively, which verifies the effectiveness of the method.