基于注意力长短期记忆深度网络的变电站上游水库水文防汛数据建模  

Modeling of Flood Data in Upstream Reservoirs of Substations Based on Attentional Long and Short-Term Memory Deep Networks

作　　者：梁允 郭志民 孟高军[2] 卢明 李哲 LIANG Yun;GUO Zhimin;MENG Gaojun;LU Ming;LI Zhe(Electric Power Research Institute of Henan Electric Power Company,Zhengzhou Henan 450052,China;School of Electric Power Engineering,Nanjing Institute of Technology,Nanjing Jiangsu 211167,China)

机构地区：[1]国网河南省电力公司电力科学研究院,河南郑州450052 [2]南京工程学院电力工程学院,江苏南京211167

出　　处：《电子器件》2025年第1期182-186,共5页Chinese Journal of Electron Devices

基　　金：国家电网有限公司总部科技项目(5500-202324180A-1-1-ZN)。

摘　　要：对变电站上游水库流入量的可靠预测是变电站防汛预报的一个关键因素。流入量的预测由于需要综合考虑气候和水文变化的影响,使其成为一项复杂的任务。开发了一种基于卷积长短期记忆的深度学习方法来实时预测排水量。这种实时预测不仅有助于水资源的有效运行,同时可以有效地监测放水的日常变化,提高运行的可靠性。通过考虑历史观测日数据中的降水量、温度、土壤含水量等信息,利用注意力长短期记忆网络异常检测算法对变电站所处地区的防汛工作进行预测。多瑙河流域观察日数据上进行的实验结果表明,所提出的方法减少了每个分析的水位测量站的误差,高水位时期的实验结果也证实所提出的方法要优于浅层模型。Reliable prediction of reservoir inflows upstream of a substation is a key factor in flood control forecasting for substations.In-flow prediction is a complex task due to the need to integrate the effects of climate and hydrological changes.A deep learning method based on convolutional long-and short-term memory is developed to predict the discharge volume in real time.This real-time prediction not only contributes to the efficient operation of water resources,but also improves the reliability of operation by effectively monitoring the daily changes in water discharge.By considering information such as precipitation,temperature,and soil moisture content from his-torical observation day data,the attentional long-and short-term memory network anomaly detection algorithm is used to predict flood control in the area where the substation is located.The results of experiments conducted on observation day data from the Danube River basin show that the proposed method reduces the error of each analyzed water level measurement station,and the experimental results for high water level periods confirm the superiority of the proposed method over the shallow model.

关 键 词：洪水预测 长短期记忆网络 注意力机制 预测模型 

分 类 号：TM933[电气工程—电力电子与电力传动]