长短时记忆神经网络模型在兴济河洪水预报中的应用

打开文本图片集

中图分类号：TV124；P338 文献标志码：A

Abstract：To enhance urban rainfallforecasting accuracyand provide more preparatory time for flood prevention，a hydrologicalpredictionresearchonthe Xingji Riverin Jinan City，ShandongProvince wasconductedusingadata-driven long short-term memory neural network（LSTM）model.Two LSTMmodelswith sequence lengths of 2.5h and 3.0h （20 were developed respectively，and their performance was optimized through comparison using root mean square error （RMSE）and Nash coeffcient.The3.Ohsequence length model wasselectedforappications in tworainfall scenarios of varying durations.Theresultsshowthatthepredictionaccuracyis relativelylowatthe main stream sourceand neartributaries.However，whenincorporating multi-dimensionalreal-timehydrologicaldata，the3.Oh sequencelengthLSTM model demonstrates significantly improvedacuracy，beteradapting tothe Xingji River’shydrologicalvariations.Additionall，the model shows superior performance in simulating subsequent rainfall events over extended periods，making it suitable for flood forecasting applications.

Keywords：hydrologic forecast；urban flood control;long short-term memory neural network model;the Xingji River

山东省济南市南部为山区，北部为平原洼地，地势南高北低。（剩余7882字）