基于时间模式注意力机制的GRU短期负荷预测

Short-term Load Prediction of GRU Neural Network Based on Temporal Pattern Attention Mechanism

针对新能源时代负荷数据随机性更强及负荷预测误差较大的问题,提出一种基于时间模式注意力机制的门控循环单元神经网络短期负荷预测方法。首先,采用自适应白噪声的完整经验模态分解对负荷数据进行处理,得到若干个频率范围不同的本征模函数。其次,通过最大互信息系数进行相关性分析筛选特征,优化输入数据的特征维度。然后,构建基于时间模式注意力机制的门控循环单元神经网络预测模型进行负荷预测,采用自回归算法优化线性特征提取能力,得到预测结果。最后,通过实例分析证明了所提模型有助于负荷预测精度的提升。

Aimed at the problem of greater randomness of load data and larger load prediction errors in the new energy era,a short-term load prediction method of gated recurrent unit neural network based on temporal pattern attention(TPA-GRU)is proposed.First,the complete ensemble empirical modal decomposition with adaptive white noise(CEEMDAN)is employed to process the load data to obtain several intrinsic mode functions(IMFs)with different fre-quency ranges.Then,the features are filtered by the correlation analysis with the maximum maximal information coeffi-cient(MIC)to optimize the feature dimension of input data.Subsequently,a TPA-GRU prediction model is constructed for load prediction.Finally,an autoregressive algorithm is used to optimize the linear feature extraction capability,and the prediction results are obtained.Through the analysis of an example,it is proved that the proposed model effectively improves the short-term load prediction accuracy.