基于Attention-GRU模型的城市燃气用气负荷预测  

Load prediction of urban gas based on Attention-GRU model

作　　者：张应辉[1] ZHANG Yinghui(Beijing Gas Group Co.Ltd.)

机构地区：[1]北京市燃气集团有限责任公司,北京市100035

出　　处：《油气储运》2022年第11期1349-1354,共6页Oil & Gas Storage and Transportation

基　　金：2020年度北京市科技计划项目“燃气用户计量端本体防护及智能安全监测关键技术研究与示范”,Z201100008120002。

摘　　要：随着天然气价格确定机制的完善和供储销体制改革的推进,城市燃气资源采购、输送通道、用户端平衡过程中面临诸多机遇和挑战,天然气消费量预测对构建城市能源安全体系显得尤为重要。依据天然气“供-储-销”计划管理体系框架,建立了基于Attention机制的GRU(Gate Recurrent Unit)城市燃气用气负荷预测模型,利用Attention机制能够捕获时间序列关键特征的优势,解决了传统时间序列预测算法对重要特征不敏感导致预测精度不高的问题;对输入特征进行统计处理、筛选,使模型聚焦于重要时间点的燃气特征信息。将新建模型应用于北京某燃气集团用气日负荷预测,并将其与常用的传统模型进行对比,结果表明:基于Attention-GRU模型的城市燃气用气负荷预测模型在预测精度上优于其他模型,Attention机制能够捕捉重要时间点局部特征,可为增强城市燃气稳定供应提供参考。With the improvement of pricing mechanism and the advancement of supply-storage-sale system reform for natural gas, many opportunities and challenges are brought in the process of urban gas procurement, transport channels and user balance. Therefore, the analysis and prediction of natural gas consumption is of great significance to the construction of urban energy security system. According to the “supply-storage-sale” plan management system of natural gas, the accurate gas prediction model for the gas supply and consumption sides was specially studied, a neural network prediction model based on time series characteristics in combination with the feature combination of Attention + GRU model was proposed,and a gas load prediction model with wider application range and higher prediction accuracy with the GRU algorithm and Attention mechanism fused was also established. In addition, the algorithm model of cyclic neural network GRU in combination with Attention was applied to urban gas load prediction for the first time, showing better prediction effect than other algorithm models, and it could provide support for enhancing the stable supply of natural gas in the region.

关 键 词：城市燃气 用气负荷 预测 神经网络 GRU模型 Attention机制 

分 类 号：TE832[石油与天然气工程—油气储运工程]