基于双隐层ANN模型的叶绿素a浓度智能预报方法  

Intelligent prediction method for Chl-a based on the artificial neural network

作　　者：何恩业[1] 杨静[1] 李尚鲁 高姗[1] HE En-ye;YANG Jing;LI Shang-lu;GAO Shan(Key Laboratory of Marine Hazards Forecasting,National Marine Environmental Forecasting Center,Ministry of Natural Resources,Beijing China 100081;Marine Monitoring and Forecasting Center of Zhejiang Province,Hangzhou 310007 China)

机构地区：[1]国家海洋环境预报中心自然资源部海洋灾害预报技术重点实验室,北京100081 [2]浙江省海洋监测预报中心,浙江杭州310007

出　　处：《海洋预报》2021年第1期44-54,共11页Marine Forecasts

基　　金：国家重点研发计划(2016YFC1401605,2016YFC1401800)。

摘　　要：利用2019年5月WZ02生态浮标监测数据,建立了两种不同隐层人工神经网络(ANN)模型的叶绿素a(Chl-a)智能预报方法,并对单隐层和双隐层模型的预测结果做了对比。结果表明:双隐层结构预测结果精度更高,泛化能力更强,一定程度上说明了深层学习比浅层学习对信息的主要特征提取能力更有优势。同时,对数据样本集合进行了系统预处理。结果显示:Chl-a浓度与溶解氧、pH、浊度和氨氮都有显著的相关性,与表层温度、盐度、亚硝氮和磷酸盐在限定时间段内的相关性不大。通过对模型预测结果的对比验证,发现数据预处理对数据质量的改进、数据挖掘执行效率和执行效果(预测结果)都起到明显的正向作用。Using the monitoring data of ecological buoy WZ02 in May 2019,two intelligent prediction methods for Chlorophyll are established based on the artificial neural network(ANN),and the prediction results of the single hidden layer model and double hidden layer model are compared.It is found that the result of the double hidden layer model is more accurate with higher generalization capability,which indicates the advantages of deep learning in extracting key characteristics compared to shallow learning.The results of data preprocessing reveals that the concentration of Chlorophyll a has significant correlation with dissolved oxygen,pH,turbidity and ammonia-nitrogen,while it shows no significant correlation with surface temperature,salinity,nitrous nitrogen and phosphate.Meanwhile,the data preprocessing plays a positive role in the improvement of data quality,data mining efficiency and prediction accuracy.

关 键 词：神经网络 智能预报 深层学习 叶绿素A 数据预处理 

分 类 号：X55[环境科学与工程—环境工程]