基于LSTMAttention网络的短期风电功率预测

提出一种基于LSTMAttention网络的短期风电功率预测方法。首先,使用LSTM网络对数值天气预测(NWP)数据的特征信息进行提取,同时采用注意力机制有效分析了模型输入与输出的相关性,从而获取了更多重要时间的整体特征；其次,使用卷积神经网络(CNN)提取NWP数据的局部特征,并引入压缩和奖惩网络(SE)模块学习特征权重,利用特征重新标定方式提高网络表示能力；最后,将局部特征和整体特征进行特征融合,通过分类器输出分类结果。利用NOAA提供的美国加利福尼亚州某风电场的数据进行案例分析,证明了所提方法的有效性。试验结果表明,与BP神经网络、自回归积分滑动平均模型(ARIMA)模型和LSTM模型相比,LSTMAttention模型具有更高的预测精度,证明了该方法的有效性。

ShortTerm Wind Power Forecasting Based on LSTMAttention Network

A shortterm wind power forecasting method based on long shortterm memoryattention (LSTMAttention) network was presented. Firstly, the LSTM network was used to extract the feature information of numerical weather prediction (NWP) data, and the attention mechanism was used to effectively analyze the correlation between input and output of the model, so as to obtain more global features of important moments. Secondly, the convolutional neural network (CNN) was used to extract the local features of NWP data, squeezeexcitation (SE) blocks were introduced to learn the feature weights, and the feature recalibration method was used to improve the network representation ability. Finally, local and global features were fused, and the classification results were output by classifier. A case study of a wind farm in California, American provided by National Oceanic and Atmospheric Administration (NOAA) was conducted to demonstrate the effectiveness of the proposed method. The experimental results showed that LSTMAttention model had higher prediction accuracy than BP neural network, autoregressive integrated moving average (ARIMA) model and LSTM model, which proved the validity of the proposed method.