基于支持向量机的珠江河口咸潮上溯预测研究

Study on Prediction of Saltwater Intrusion Based on Support Vector Machine in the Pearl River Estuary

受径流、潮汐、地形、风向等多重因素影响，珠江河口区咸潮上溯现象异常复杂。本文运用支持向量机回归（SVR）与分类（SVC）联合建模的方法，综合考虑上游径流和河口区潮汐因素，构建珠江河口区磨刀门水道咸潮上溯预测模型，并采用动态反馈机制分别开展了磨刀门水道日均咸度和超标历时预测。研究结果表明：基于支持向量机的咸潮上溯模型，具有模型简单精度高，样本容量要求少，自学习与联想记忆能力强的特点，能较好模拟预测多重因素影响的复杂河口区咸潮上溯现象；磨刀门水道上游三水马口前三日流量、河口区前三日三灶站的平均潮差过程与咸潮上溯具有较好时间相关性。

Effected by multiple factors such as upstream runoff, tide, topography, and wind, the saltwater in- trusion of Pearl River Estuary displays extreme complexity. In this paper, a hybrid method based on Support Vector Machine CSVR） and Classification （SVC） is used to build a daily average salinity prediction model in the Pearl River Estuary which has considered the variation of upstream runoff and tide. The result shows that the model for saltwater intrusion prediction has high accuracy and only requires a few samples. Furthermore, the model has a good ability of self-learning and asso- ciative memory which can well simulate the complex saltwater intrusions of this estuary. The findings of this study also show that the previous average three-day runoff from upstream and the previous three-day tidal range from outlet have a good time correlation with saltwater intrusion in this estuary.