欢迎访问《渔业研究》官方网站,今天是 分享到:

›› 2020, Vol. 42 ›› Issue (1): 1-9.

• 论文与报告 •    下一篇

三十六脚湖叶绿素a浓度人工神经网络模型演算研究

许阳春1,覃苗2,苏玉萍1,林晓萍3,苏金洙3   

  1. 1. 福建师范大学
    2. 福建师范大学环境 科学与工程学院
    3.
  • 收稿日期:2019-09-16 修回日期:2019-10-18 出版日期:2020-02-25 发布日期:2020-03-02
  • 通讯作者: 许阳春
  • 基金资助:
    国家重点研发计划;国家自然科学基金;福建省科技厅高校产学合作项目;福州市科技局项目

Study on using BP artificial neural network model for chlorophyll-a concentration in Pingtan Thirty-six Feet Lake

  1. 1. Fujian Normal University
    2.
    3. undefined
  • Received:2019-09-16 Revised:2019-10-18 Online:2020-02-25 Published:2020-03-02
  • Contact: Yang-chun XU
  • Supported by:
    National Key R&D Program of China; National Natural Science Foundation of China;University-Industry Cooperation Program of Fujian Science and Technology Department;Fuzhou Science and Technology Bureau Program

摘要: 本文分析了平潭三十六脚湖2016年1月至2017年5月气象和水质历史数据,以环境因子作为输入参数,叶绿素a浓度作为输出参数,构建了BP人工神经网络模型。从历史样本数据中随机抽取80%的数据进行模型演算,剩下20%的数据作为测试数据进行检验。结果表明:以气温、电导率、水温3个指标为输入因子时,模型输出的叶绿素a浓度和监测数据的拟合度达到R2=0.97,RMSE=0.05 μg/L、RSR=0.17,误差较小。2019年3月13日至4月26日对三十六脚湖进行5次采样,将实测的叶绿素a浓度值与模型演算值进行对比分析,发现其标准偏差比RSR为0.24,实测值与演算值的偏离程度较小,精度达到期望值。该模型有望用于平潭三十六脚湖湖区叶绿素a浓度预测和水华预警,为水体富营养化防控提供参考。

关键词: 叶绿素a浓度, 模型演算, BP人工神经网络, 三十六脚湖, chlorophyll-a, model calculus, BP artificial neural network, Thirty-six Feet Lake

Abstract: We collected the meteorological and water quality data of Pingtan Thirty-six Feet Lake from January 2016 to May 2017 and developed a model based on the BP artificial neural network. Environmental factors were used as input indicators to run the model, while the chlorophyll-a concentration was used as an output index. 80% historical data were randomly extracted from the sample data for model calculating, and others were used for model verification. The results showed that the fitting degree of chlorophyll-a concentration and monitoring data reached R2=0.97, RMSE=0.05 μg/L, RSR=0.17, and the error was small when the temperature, conductivity and water temperature were taken as input factors. We sampled the lake for five times from March 13th to April 26th, 2019 by comparing the measured chlorophyll-a concentration with the predicted chlorophyll-a concentration. It showed that the RSR was 0.24, the deviation between measured and predicted values was small. The results showed that the model was expected to be applied to the prediction of chlorophyll-a concentration and early warning of water blooms in the area of Pingtan Thirty-six Feet Lake, which would provide a reference for the prevention and control of eutrophication.