基于BP人工神经网络的北江水体中叶绿素a浓度预测模型研究  

Study on Concentration Prediction Model of Chlorophyll a in Beijiang River Base on BP Artificial Neural Network

作　　者：黄成 吴剑 丁炜炜 吴淇钰 邱劼婷 张祺悦 王兴 HUANG Cheng;WU Jian;DING Wei-wei;WU Qi-yu;QIU Jie-ting;ZHANG Qi-yue;WANG Xing(Shaoguan Ecological&Environmental Monitoring Center Station of Guangdong Province,Shaoguan,Guangdong 512026,China)

机构地区：[1]广东省韶关生态环境监测中心站,广东韶关512026

出　　处：《四川环境》2025年第1期109-115,共7页Sichuan Environment

基　　金：韶关市科技计划基金资助项目(210725104530665)。

摘　　要：在北江选取了4个自动实时监测水质的网点作为研究断面,利用水站在2021年藻类水华期间以及2022年至2023年跟踪期间实时监测数据和人工调查水生态数据,通过主成分分析和北江藻类水华机理分析,以叶绿素a输出作为表征藻类水华生物量,设置了3组不同参数组合进行BP模型演算,演算数据共155组,随机选取80%数据作为训练样本,其余进行模型验证。模型演算效果显示水温、pH、COD_(Mn)组合为最佳输入组合,BP模型误差较小(均方根误差为6.74μg/L,平均绝对误差为9.26μg/L),演算结果精度较高(可决系数R^(2)=0.892)。使用训练好的模型,输入水站在线监测数据对叶绿素a进行预测,预测值和实测值的均方根误差降至1.96μg/L。结果表明,水温、pH、COD_(Mn)对叶绿素a浓度预测效果好,此模型可较好地为北江藻类水华预测预警和防控工作提供技术支持。Four automatic and real-time water quality monitoring points in the Beijiang River were selected as research sections.The real-time monitoring data from these water quality monitoring stations during the algal bloom period in 2021 and the tracking periods from 2022 to 2023,as well as water ecology data from artificial investigation were used to carry out a principal component analysis and mechanism analysis of algal blooms in the Beijiang River.Using the output of chlorophyll a as a characterization of algal bloom biomass,three different parameter combinations were set up for BP model calculation.A total of 155 sets of calculation data were obtained,and 80% of the data were randomly selected as training samples,while the rest were subjected to model validation.The model calculation showed that the combination of water temperature,pH,and COD_(Mn)is the optimal input combination,while the BP model has a small error(root mean square error of 6.74μg/L,with an average absolute error of 9.26μg/L),and the accuracy of the calculation results was very high(the coefficient of determinability R^(2)=0.892).With the trained model,the online monitoring data of the water quality monitoring stations were inputted to predict chlorophyll a,and the root mean square error between the predicted and measured values was reduced to 1.96μg/L.The results indicated that water temperature,pH,and COD_(Mn)had good predictive effects on chlorophyll a concentration,and the model can provide technical support for prediction,warning,and prevention of algal blooms in the Beijiang River.

关 键 词：BP人工神经网络 藻类水华 叶绿素A浓度 预测 北江 

分 类 号：X832[环境科学与工程—环境工程]