基于MHA-BiGRU的内河航道水位预测分析

Water level prediction of inland waterways based on MHA-BiGRU

扫码查看

原文链接

万方数据

中文摘要：针对山区内河航道水位预测技术难题,分析了影响航道水位预测的环境因素和技术难点,总结了当前航道水位预测模型的技术方法,提出一种新型的MHA-BiGRU航道水位预测模型,将多头注意力机制引入GRU模型,实现了模型对航道水位序列数据时间和空间等重要因素的特征权重划分,使模型聚焦影响航道水位变化的关键因素.以乌江下游航道为研究对象,通过建设水位和流速实时动态监测站,建立监测数据集,选取平均绝对误差(MAE)、纳什效率系数(NSE)和均方根误差(RMSE)等参数作为评价指标对该模型进行验证.结果表明,模型通过多头注意力机制和双向循环神经网络的应用,提升了航道水位预测性能;与传统的LSTM和GRU等经典时序预测模型相比,该模型具有更好的鲁棒性和更高的精度.将模型嵌入系统平台中进行示范应用,实现了航道水位的实时动态监测和中短期预测,具有较高的工程应用价值.

外文摘要：Aiming at the technical problems of inland waterway level prediction in mountainous area,the environmental fac-tors and technical difficulties affecting waterway water level prediction were analyzed,the current technical methods of waterway water level prediction model were summarized,and a new waterway water level prediction model of multi-head at-tention-bidirectional gated recurrent unit(MHA-BiGRU)was proposed,the multi-head attention mechanism was introduced into GRU model,and the characteristic weights of important factors such as time and space of waterway water level series data were divided by the model,so that the model focused on the key factors affecting waterway water level change.Taking the downstream of Wujiang River as the research object,the model realized the establishment of monitoring data set by building a real-time dynamic monitoring station for water level and flow velocity.The parameters such as MAE,RMSE and NSE were selected as evaluation indexes to verify the proposed model.Results show that the model improves the performance of waterway water level prediction by the application of multi-head attention mechanism and bidirectional cyclic neural net-work.Compared with traditional classical time series predic-tion models such as LSTM and GRU,the model has better ro-bustness and higher accuracy.The model is embedded into the system platform for demonstration application,and the real-time dynamic monitoring and medium-and short-term predic-tion of waterway water level are realized,with higher engi-neering application value.

外文关键词：

inland waterwaywater level predictionwater-way safety supervisionmulti-head attention(MHA)gated recurrent unit(GRU)

作者：

马瑞鑫、尹勇、鲍可馨、汪永超

展开 >

作者单位：

大连海事大学航海动态仿真和控制实验室,辽宁大连 116026

交通运输部天津水运工程科学研究院,天津 300456

关键词：

内河航道水位预测航道安全监管多头注意力机制(MHA) 门控循环单元(GRU)

基金：

国家重点研发计划项目广西科技计划项目重点研发计划中央级公益性科研院所科研创新基金资助项目

项目编号：

2023YFB2603804桂科AB22080106TKS20230203

出版年：

2024

DOI：

10.16411/j.cnki.issn1006-7736.2024.01.006

大连海事大学学报

大连海事大学

大连海事大学学报

CSTPCD北大核心

影响因子：0.469

ISSN：1006-7736

年,卷(期)：2024.50(1)

参考文献量26