EMD-Attention-GRU天然气管网流量组合预测模型

为解决传统时间序列预测方法在天然气管网流量预测中存在的不足，提出一种基于经验模态分解（Empirical Mode Decomposition，EMD）、注意力机制（Attention）和门控循环单元（Gated Recurrent Unit，GRU）的组合模型。该模型利用经验模态分解得到的原始天然气管网流量时间序列分量代替原始天然气管网流量数据，再将得到的本征模态函数分量输入GRU神经网络，利用在网络中集成的注意力机制计算不同时刻的注意力概率权重，最后在网络中学习并预测天然气管网流量时间序列。某天然气管网实例验证结果表明：EMD-Attention-GRU组合模型在预测天然气管网流量方面表现出良好的性能，能够捕捉到复杂的非线性关系，相比单一GRU模型和Attention-GRU模型，其预测结果的平均绝对百分比误差指标分别降低6.29%和5.17%。由此说明，与传统时间序列预测方法相比，EMD-Attention-GRU组合模型能够更好应对天然气管网流量的复杂性和动态特征，具有推广应用价值。

A combined forecast model is proposed to address the limitations of using traditional time series forecasting for natural gas pipeline flow. The model integrates Empirical Mode Decomposition (EMD), Attention mechanism, and Gated Recurrent Unit (GRU) to enhance the accuracy and stability of natural gas flow predictions. The model utilizes Empirical Mode Decomposition and Attention mechanism to decompose the original data and assign different attention weights to each moment's data, thereby enhancing the accuracy of the predictions. The introduced EMD-Attention-GRU combination model demonstrates remarkable performance in forecasting natural gas flow, successfully capturing intricate non-linear patterns. In comparison to conventional time series forecasting methods, the combination model exhibits superior adaptability in addressing the complexities and dynamic features of natural gas flow. During the instance verification, the average absolute percentage error of the combination model's predictions outperforms single GRU and Attention-GRU models by 6.29% and 5.17%, respectively.