摘要
Abstract
Accurate long-term wind power prediction is crucial to the stable operation of the power grid system.Traditional forecasting methods are not effective in handling long series prediction.Recent studies had shown that Informer model has achieved good results in the field of long series prediction.However,the model still needed to be improved in capturing the local features of the data and dealing with the stacking of network layers,so a long-term wind power prediction method based on KCR-Informer was proposed in this paper.Firstly,the impact of meteorological data on wind power was analyzed,and Kalman filter was used to smooth wind power meteorological data to reduce the impact of noise on data.Subsequently,a wind turbine power prediction model was established based on Informer,enabling long-term power prediction using both meteorological data and historical power data.Moreover,convolutional neural networks and residual connection modules were introduced to enhance the Informer model,so that the model could better capture the local features,accelerated the model convergence,and solved the problem of model network degradation.The computational results demonstrated that,compared with the long short-term memory(LSTM)algorithm,Transformer algorithm,and Informer algorithm,the proposed method in this paper achieved a reduction in the mean absolute error(MAE)ranging from 5.7%to 30%and a reduction in the mean square error(MSE)ranging from 8.3%to 35%for different prediction horizons.This significant improvement validated the effectiveness of the proposed model in enhancing long-term wind power prediction accuracy.关键词
长期风电功率预测/卡尔曼滤波器/Informer模型/卷积神经网络/残差连接Key words
long-term wind power prediction/Kalman filter/Informer model/convolutional neural network/residual connection分类
信息技术与安全科学
