基于卷积长短期记忆神经网络的短期风功率预测

SHORT-TERM WIND POWER PREDICTION BASED ON CONVOLUTIONAL LONG-SHORT-TERM MEMORY NEURAL NETWORKS

提出一种基于卷积长短期记忆神经网络(CNN-LSTM)的短期风功率预测模型。该模型以风电场风功率历史数据以及风速风向等数值天气预报(NWP)数据为输入对风功率进行预测。首先,利用主成分分析法(PCA)对原始多维气象数据变量进行预处理,然后将处理过的气象数据和历史风功率数据通过卷积网络实现对数据的特征提取和进一步的数据降维,再通过长短期记忆网络实现对数据的拟合,并在神经网络的训练过程中引入DropConnect技术减小模型中的过拟合现象,最终实现风功率的精确预测。以中国西北某风电场的实测数据进行验证,结果表明所提方法能有效对风功率进行预测,较BP神经网络和支持向量机(SVM)有更高的预测精度。

This paper proposes a new short-term wind power prediction model based on convolution long-short-term memory neural networks(CNN-LSTM),which uses wind power historical data and numerical weather prediction(NWP)data including wind speed and wind direction as the input information.Firstly,the original multi-dimensional input variables are preprocessed by principal component analysis(PCA),and the principal components of the input variables are selected as the input of the neural networks.Then the processed data is extracted by the convolutional networks,and is next fitted through the long-short-term memory networks.The DropConnect technique is introduced in the training process of the neural networks to reduce the over-fitting phenomenon in the model.Finally,accurate wind power prediction is achieved.This model is validated by predicting a real wind farm output in Northwest China.The results show that the proposed method can effectively predict wind power and has higher prediction accuracy than that by BP neural networks and support vector machine(SVM).