基于机器学习的珠江河口咸潮上溯实时预报研究

打开文本图片集

中图分类号：TV11 文献标识码：A 文章编号：1001-9235（2025）11-0001-08

Real-time Forecasting of Tidal Saltwater Intrusion in the Pearl River Estuary Based on Machine Learning

YIJingjing'，LIUDawei’，ZHUYuke²，LIUBingjun 2，3

（1.Guangdong ProvincialHydrological Bureau，Foshan HydrologicalSub-bureau，Foshan 5280o0，China; 2.SchoolfCivil Enginering，Sun Yat-sen University，Zhuhai519085，China; 3.ResearchCenterfor WaterResourcesandEnvironment，SunYatsen University，Guangzhou510275，China）

Abstract：IntensifiedglobalclimatechangeadhumanactivitisleadtoincreasinglyseveretidalsaltwaterintrusioninthePearlRiver Estuary，andthewatersuplysecurityofcoastalcitiesisundersignificantthreat.Thisstudyemploedlongshort-termmemory （LSTM）andgatedrecurentunit（GRU）networks toforecastandvalidatehourlysalnitydataat thePinggangStationintheModaomen Waterway from 2019 to 2023. The results indicate： ① Both the LSTM and GRU models demonstrate strong performance in forecasting tidalsaltwaterintrusionComparedtoeLSTodel，teGUodeleitsgerfreastinguacalerprdictioo， andfaster computational speed， with its performance advantages being more pronounced in short-term forecasts. ② The GRU model achievesaforecastingaccuracyofaboveO.8forfuture1～24hours，withtheaccuracyforfuture1～6hoursgenerallyreaching0.9. KeyWords： tidal saltwater intrusion forecasting; LSTM model; GRU model; deep learning; Pearl River Estuary

受全球气候变迁与人类活动强度叠加影响，珠江河口咸潮入侵现象呈现显著加剧态势，对沿岸城市群供水安全构成严峻挑战 ⌊1-2⌋ 。（剩余9466字）