基于注意力机制的卷积神经网络-长短期记忆网络的短期风电功率预测

Short-Term Wind Power Forecasting Based on Attention Mechanism of CNN-LSTM

为了提高风电功率的预测精度,针对风电数据间歇性与时序性的特点,提出了一种基于注意力机制的卷积神经网络-长短期记忆(convolutional neural networks-long short-term memory,CNN-LSTM)网络预测模型。首先利用CNN提取风电数据动态变化的多维特征,然后将特征向量构造成时序形式并作为LSTM网络的输入,最后使用注意力机制进行优化,通过赋予LSTM网络隐含层不同的权重,增强重要信息的作用,完成风电功率预测。采用国内某风电场的风电数据进行实验,结果表明该模型比支持向量机、LSTM模型、CNN-LSTM模型具有更好的预测精度。

To improve the prediction accuracy of wind power,in view of the characteristic of intermittent and time-sequence of wind power data,a convolutional neural networks-long shortterm memory(abbr.CNN-LSTM)prediction model based on attention mechanism was proposed.Firstly,by use of CNN the multi-dimension feature of dynamic variation of wind power data was extracted.Secondly,the feature vector was constructed into time sequence form and was used as the input of the LSTM network.Finally,the attention mechanism was used to optimize,then by means of endowing different weights to the hidden layer of LSTM to enhance the role of important information,so that the wind power prediction was completed.The wind power data of a certain domestic wind farm is utilized for simulation and simulation results show that the prediction accuracy by the proposed model is higher than those from support vector machine,LSTM model and CNN-LSTM model.