基于改进长期循环卷积神经网络的海上风电功率预测

Offshore Wind Power Prediction Based on Improved Long-term Recurrent Convolutional Neural Network

准确的风电功率预测对海上风电安全并网具有重要意义。不同于陆地,海上具有气象因素复杂、风电出力波动显著等特点,使得海上风电功率预测精度难以满足工程实际要求。针对以上问题,文中提出一种基于改进长期循环卷积神经网络(LRCN)的预测模型,用于超短期海上风电功率预测。首先,采用改进LRCN进行初步功率预测,即构建多卷积通道分别提取不同层次变量的时序特征,并通过具有前瞻性的改进Adam优化器提升网络收敛效果。其次,利用摇摆窗算法与波动特征聚类识别预测时段的出力波动类型。再次,针对不同的波动类型建立对应的误差修正模型,并输入经Xgboost算法筛选出的强相关特征因子,实现误差修正。最后,采用实际海上风电场数据进行实验,其结果表明所提方法能够有效预测超短期海上风电功率,且预测精度高于多种传统预测模型。

Accurate wind power prediction is of great significance to the safe connection of offshore wind power for the grid.Different from the land,the sea has the characteristics of complicated meteorological factors and significant fluctuation of wind power output,which makes the prediction accuracy of offshore wind power difficult to meet the practical engineering requirements.Aiming at the above problems,this paper proposes a prediction model based on the improved long-term recurrent convolutional neural network(LRCN)for ultra-short-term offshore wind power prediction.Firstly,the improved LRCN is used for preliminary power prediction,that is,the multi-convolution channel is constructed to extract the time series characteristics of variables at different layers,and the network convergence effect is improved by the forward-looking improved Adam optimizer.Secondly,the swing window algorithm and the clustering of fluctuation characteristics are used to classify the types of output fluctuation in the predicted period.Thirdly,error correction models are established for different fluctuation types,and the strongly correlated feature factors screened by the Xgboost algorithm are input to achieve error correction.Finally,experiments with data of actual offshore wind farm is put forward,and the results show that the proposed method can effectively predict the ultra-short-term offshore wind power,and the prediction accuracy is higher than that of traditional prediction models.