基于KCR-Informer的长期风电功率预测研究

Research on Long-term Wind Power Prediction Based on KCR-Informer

准确的长期风电功率预测对电网系统稳定运行至关重要,传统预测方法在处理长序列预测时效果并不理想,近期研究表明Informer模型在长序列预测领域取得良好效果。然而,该模型在捕捉数据的局部特征以及处理网络层数堆叠问题上还有待改进。文章提出一种基于卡尔曼滤波器-卷积神经网络-残差网络-Informer(Kalman filter-convolutional neural network-residual network-informer,KCR-Informer)模型的长期风电功率预测方法,首先分析气象数据对风电功率的影响,使用卡尔曼滤波器对风电气象数据进行数据平滑处理,以减轻噪声对数据的影响,然后基于Informer模型建立风电功率预测模型,根据气象数据以及历史功率数据进行长期功率预测;在此基础上,引入卷积神经网络和残差连接模块,使模型能够更好的捕捉到局部特征,同时加快模型收敛,解决模型网络退化问题。算例的结果表明,与长短期记忆网络(long short-term memory,LSTM)算法、Transformer算法、Informer算法相比,文章方法在不同预测步长下的平均绝对误差(mean absolute error,MAE)降低5.7%~30%,均方误差(mean square error,MSE)降低8.3%~35%,长期风功率预测的精度得到提升,验证了模型的有效性。

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.