摘要
为提升水利设施监测站点周边水位预测的准确性,设计基于双向LSTM神经网络的站点周边水位预测系统;系统硬件部分设计了周边水系查询体系与水位记录装置;系统软件根据初始参数定义结果建立LSTM神经网络布局模型,设计双向LSTM解码器,其连接闭环能够有效地提高模型的预测性能和稳定性;采集水位数据并进行清洗处理,利用清洗后的数据对象建立一维水动力模型,计算水系糙率,确定流量与延时时间的数值关系,将上述参数作为输入值对双向LSTM模型进行训练,实现水位信息的预测;实验结果表明,在实验水系区域内,所提方法5月份、6月份的水位记录数据与原始水位数据之间的差值始终为零,拟合误差也为零;而对比方法中基于DWT-LSTM的水位预测模型的5月份、6月份的水位差值分别为1.9 m、1.1 m;抽水蓄能引水型水位监测系统的5月份、6月份的水位差值分别为3.0 m、2.4 m。
To improve the accuracy of water level prediction around water conservancy facility monitoring stations,a water level prediction system for station surroundings based on bidirectional long short-term memory(LSTM)neural network is designed.The hardware part of the system is designed with the peripheral water system query system and the water level recording device;The system software establishes the LSTM neural network layout model based on the initial parameter definition results,designs the bidirectional LSTM decoder,and its connection loop can effectively improve the predictive performance and stability of the model.The system collects the water level data and performs cleaning processing.The cleaned data object is used to establish the one-dimensional hydrodynamic model,calculate the roughness of the water system,determine the numerical relationship between flow rate and delay time,and train the bidirectional LSTM model with the above parameters as the input values to achieve the prediction of water level information.Experimental results show that within the experimental water system area,the difference between the water level recorded data in May and June using the proposed method and the original water level data is always zero,and the fitting error is also zero.The water level difference between May and June based on the DWT-LSTM water level prediction model in the comparison method is 1.9 m and 1.1 m,respectively;The water level difference between May and June for the pumped storage and diversion water level monitoring system is 3.0 m and 2.4 m,respectively.
作者
姚晔
许锡伟
管剑波
葛旭初
YAO Ye;XU Xiwei;GUAN Jianbo;GE Xuchu(Ningbo Rail Transit Group Co.,Ltd.,Ningbo 315000,China)
出处
《计算机测量与控制》
2024年第11期18-24,33,共8页
Computer Measurement &Control
关键词
双向LSTM神经网络
水位预测
水系查询
数据清洗
水动力模型
水系糙率
bidirectional LSTM neural network
water level prediction
water system inquiry
data cleaning
hydrodynamic model
drainage roughness
