耦合新安江模型与LSTM模型的赣江流域径流模拟研究
Runoff Simulation in the Ganjiang River Basin Based on the Xin'anjiang Model Coupled with the LSTM
邹佳成 1黄监初 1杨丽琳 2顾雯叶3
作者信息
- 1. 赣江下游水文水资源监测中心,江西 宜春 336028
- 2. 山东省水利勘测设计院有限公司,山东 济南 250013
- 3. 中交第三航务工程勘察设计院有限公司,上海 200032
- 折叠
摘要
准确的径流模拟对流域水资源优化配置、防汛抗旱具有重要作用.为此,以赣江流域为例,构建了新安江模型与 LSTM的耦合模型(XAJ-LSTM),对比分析了 XAJ-LSTM、新安江模型和 LSTM模型的径流模拟差异,并评估了汛期划分对径流模拟精度的影响.结果表明,在赣江流域,LSTM模型的最佳神经元参数为第一层 18 个,第二层 36 个;LSTM模型的径流模拟效果要优于新安江模型,且利用新安江模型对 LSTM模型进行物理约束后,RRMSE 的降幅达 11%;考虑汛期划分能够提升 XAJ-LSTM模型对汛期与非汛期径流的模拟精度,且在汛期的改善效果更明显,RRMSE 减幅达 18%.研究成果可为赣江流域径流模拟与预报提供参考.
Abstract
Accurate streamflow simulation plays an important role in the optimal allocation of water resources,flood control and drought relief.In this study,taking Ganjiang River Basin for an example,a coupled XAJ-LSTM model of the Xin'anjiang(XAJ)model and LSTM model was constructed,and the differences in streamflow simulation between the coupled model,the XAJ model and the LSTM model are compared and analyzed.The impact of flood season division on the streamflow simulation accuracy was evaluated.The results show that the optimal neuron combination of LSTM model is 18 neurons in the first layer and 36 neurons in the second layer.Compared with the XAJ model,LSTM model has high simulation accuracy;Furthermore,the LSTM model coupled physical constraints based on the XAJ model achieves ap-proximate 11%lower of the root mean squared error RRMSE.Considering the flood season division can improve stream-flow simulation accuracy of the XAJ-LSTM model,the improvement is more significant in the flood season,with the RRMSE reduced by 18%.This study can provide technique support for the streamflow simulation and prediction for the Ganjiang River Basin.
关键词
径流模拟/长短期记忆神经网络/新安江模型/汛期划分Key words
streamflow simulation/LSTM/Xin'anjiang model/flood season division引用本文复制引用
基金项目
江西省"科技+水利"联合计划项目(2022KSG01006)
出版年
2024
NETL
NSTL
万方数据