基于LSTM模型的海洋水质预测  

Marine Water Quality Prediction Based on LSTM Model

作　　者：李彦杰 贺鹏飞[1] 冯巍巍 刘巧利 杨信志 LI Yanjie;HE Pengfei;FENG Weiwei;LIU Qiaoli;YANG Xinzhi(Photoelectricity Information Science Technology Institute,Yantai University,Yantai 264005;Yantai Institute of Coastal Zone Research,Chinese Academy of Sciences,Yantai 264003)

机构地区：[1]烟台大学光电信息科学技术学院,烟台264005 [2]中国科学院海岸带研究所,烟台264003

出　　处：《计算机与数字工程》2020年第2期437-441,共5页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61202399);国家级大学生创新创业训练计划项目(编号:201811066004)资助。

摘　　要：海洋水质实时预测和实时数据处理技术有利于充分利用海洋资源,发展海洋经济,解决沿海水域海水污染和水质监管问题。论文运用深度学习的长短时记忆网络(Long Short Term Memory Networks,LSTM)算法对不同时间段内采集到的海洋水质数据进行分析建模,以实现对未来海水水质的预测。与传统的支持向量回归(Support Vector Regression,SVR)算法相比,LSTM取得了更好的拟合效果,拟合优度达到0.9554,平均绝对误差为0.0117,能很好地实现对海水水质的全天候预测,从而有效地监管海洋水质变化情况,提高海洋污染预警及海洋生态保护能力。Real-time prediction and data processing of marine water quality are conducive to making full use of marine resources. It is also a powerful tool for developing the marine economy,solving the seawater pollution and water quality supervision problems in coastal waters. In this paper,Long Short Term Memory Networks(LSTM)algorithm for deep learning is used to conduct training modeling for Marine water quality data collected in different time periods. Compared with the traditional Support Vector Regression(SVR)algorithm,LSTM achieves a better fitting effect with a goodness-of-fit of 0.9554 and an average absolute error of 0.0117. The proposed algorithm can well achieve all-weather prediction of marine water quality,so as to effectively monitor the marine water quality changes. Furthermore,it can improve the marine pollution warning and capacity of marine ecological protection.

关 键 词：海洋水质预测 机器学习算法 长短时记忆网络 支持向量回归 水质预测 污染预警 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]