基于Informer模型的开都河流域径流预测  

Runoff prediction in the Kaidu River Basin based on Informer model

作　　者：罗鑫 Luo Xin(China Energy Group Kaidu River Basin Hydropower Development Co.,Ltd.,Korla831000,Xinjiang,China)

机构地区：[1]国家能源集团开都河流域水电开发有限公司,新疆库尔勒831000

出　　处：《吉林水利》2024年第11期58-64,共7页Jilin Water Resources

摘　　要：为提高径流预测的准确性,本文通过应用长短期记忆(Long Short-Term Memory,LSTM)和Informer模型,对开都河流域察汗乌苏水文站2010年6月至2023年6月的逐日径流数据进行模拟。结果表明Informer模型的模拟性能最优,验证期的NSE达到0.96,而LSTM模型在径流峰值处的模拟存在低估。因此,基于Informer模型建立了开都河流域的单变量和多变量径流预测模型。结果表明,Informer模型在开都河流域察汗乌苏水文站具有较好的适用性。单变量和多变量预测模型在预测步长为1-3d时精度最高,R2>0.9,RMSE<15,MAE<10。对比单变量预测模型与考虑降水和气温的多变量预测模型的预测结果,结果表明,随着预见期逐渐增加,单变量预测模型性能更好。研究结果验证了Informer模型在径流预测领域的有效性,为研究区的水资源管理和防洪减灾等决策部门提供了一定参考。In order to improve the accuracy of runoff prediction,this paper simulates the daily runoff data of the Chahanwusu Hydrological Station in the Kaidu River Basin from June 2010 to June 2023 by applying the Long Short-Term Memory(LSTM)and Informer model.The results show that the simulation performance of the Informer model is the best,and the NSE reaches 0.96 during the validation period,while the simulation of the LSTM model at the peak of runoff is underestimated.Therefore,univariate and multivariate runoff prediction models for the Kaidu River Basin were established based on the Informer model.The results show that the Informer model has good applicability in the Chahanwusu hydrological station in the Kaidu River Basin.Univariate and multivariate prediction models have the highest accuracy when the prediction step size is 1-3d,R2>0.9,RMSE<15,and MAE<10.The comparison of the prediction results of the univariate prediction model and the multivariate prediction model considering precipitation and temperature shows that the univariate prediction model performs better with the gradual increase of the prediction period.The results of this study verify the effectiveness of the Informer model in the field of runoff prediction,and provide a certain reference for decision-making departments such as water resources management and flood control and disaster reduction in the study area.

关 键 词：深度学习方法 LSTM模型 Informer模型 日径流预测 开都河流域 

分 类 号：P338[天文地球—水文科学]