基于MIC-BP神经网络溶解氧预测模型的构建和评价

doi:10.14012/j.cnki.fjsc.2023.04.001

摘要/Abstract

摘要：

溶解氧(DO)是影响水生生物生长和水环境健康的重要环境因子,对DO精准预测有利于水产养殖业的健康发展。本研究根据福建闽江水口库区水质在线浮标SK11、SK18站位2022年1月到6月的水质数据和气象数据,分别运用BP神经网络预测模型和MIC-BP神经网络预测模型进行机器学习,给出预测结果,同时对两种DO预测模型的预测结果进行比较验证。结果表明:经过最大信息系数(MIC)的识别和筛选,13项输入因子中与DO相关性较大的因子有pH、水温、叶绿素a、电导率、浊度、氨氮浓度和亚硝酸盐氮浓度等7项;混合MIC-BP神经网络模型的效果明显优于独立的BP神经网络模型,候选因子经过MIC的识别和筛选后可以明显增加模型的性能,表现为:在SK11站位,MIC-BP神经网络模型的性能相对于独立BP神经网络模型,MAE降低约29.29%,RMSE降低约60.09%,NSE增加27.63%;在SK18站位,MIC-BP神经网络模型的性能相对于独立BP神经网络模型,MAE降低约17.16%,RMSE降低约16.23%,NSE增加12.77%。

关键词: 溶解氧预测, 最大信息系数(MIC), 机器学习, BP神经网络

Abstract:

Dissolved oxygen is an important environmental factor that affects the growth of aquatic organisms and water environment. Accurate prediction of dissolved oxygen is beneficial to the healthy development of aquaculture. This study was based on the water quality data and meteorological data of online buoys SK11 and SK18 in Shuikou reservoir area of Minjiang River in Fujian from January to June, 2022. Then, back propagation(BP) neural network prediction model and MIC-BP neural network measurement model were used for machine learning, and the prediction results are given. At the same time, the prediction results of the two dissolved oxygen prediction models were compared and verified. The results showed that after the identification and screening of MIC (Maximum information coefficient), among the 13 input factors, the factors that had great correlation with dissolved oxygen include pH, water temperature, chlorophyll, electrical conductivity, turbidity, ammonia nitrogen concentration and nitrite nitrogen concentration. The effect of the mixed MIC-BP neural network model was obviously better than that of the independent BP neural network model. After the candidate factors were identified and screened by MIC, the performance of the model could be obviously improved. Compared with the independent BP neural network model, the results showed that the performance of MIC-BP neural network model at SK11 station decreased by 29.29%, RMSE decreased by 60.09%,and NSE increased by 27.63%,respectively. At station SK18, MAE decreased by 17.16%, RMSE decreased by 16.23%, and NSE increased by 12.77%,respectively.

Key words: dissolved oxygen prediction, maximum information coefficient (MIC), machine learning, back propagation(BP) neural network

中图分类号:

TP183

陈梦云. 基于MIC-BP神经网络溶解氧预测模型的构建和评价[J]. 渔业研究, 2023, 45(4): 317-330.

CHEN Mengyun. Construction and evaluation of dissolved oxygen prediction model based on MIC-BP neural network[J]. Journal of Fisheries Research, 2023, 45(4): 317-330.

图/表 15

图1 研究区域在线浮标站位

Fig.1 Online buoy station in the study area

图2 模型构建流程

Fig.2 The procedures of model construction

图3 BP 神经网络训练过程

Fig.3 BP neural network training process

图4 SK11站位水质参数1—6月变化特征

Fig.4 Variation characteristics of water quality parameters at SK11 station from January to June

图5 SK18 站位水质参数1—6月变化特征

Fig.5 Variation characteristics of water quality parameters at SK18 station from January to June

图6 SK11、SK18站位各候选因子的MIC值

Fig.6 MIC values of candidate factors in SK11 and SK18 stations

表1 SK11、SK18站位模拟效果

Tab.1 Simulation effect of SK11 and SK18 stations

浮标站位 Buoy station	数据频次/h Frequency of data	模型 Type of model	R²	MAE	RMSE	NSE
SK11	1	独立BP	0.97	0.99	2.33	0.55
SK11	1	MIC-BP	0.97	0.70	0.93	0.76
SK18	1	独立BP	0.97	1.69	3.08	0.41
SK18	1	MIC-BP	0.97	1.40	2.58	0.47

图7 SK11 站位独立BP神经网络模型拟合图

Fig.7 The fitting diagram of independent BP neural network model at SK11 station

图8 SK11 站位独立BP神经网络模型的决定系数

Fig.8 The decision coefficient of independent BP neural network model at SK11 station

图9 SK11 站位混合MIC-BP神经网络模型拟合图

Fig.9 The fitting diagram of mixed MIC-BP neural network model at SK11 station

图10 SK11 站位混合MIC-BP神经网络模型的决定系数

Fig.10 The decision coefficient of mixed MIC-BP neural network model at SK11 station

图11 SK18 站位独立BP神经网络模型拟合图

Fig.11 The fitting diagram of independent BP neural network model at SK18 station

图12 SK18 站位独立BP神经网络模型的决定系数

Fig.12 The decision coefficient of independent BP neural network model at SK18 station

图13 SK18 站位混合MIC-BP神经网络模型拟合图

Fig.13 The fitting diagram of mixed MIC-BP neural network model at SK18 station

图14 SK18 站位混合MIC-BP 神经网络模型的决定系数

Fig.14 The decision coefficient of mixed MIC-BP neural network model at SK18 station

参考文献 16

[1]	江天棋, 张扬, 姜亚洲, 等. 不同溶解氧水平下厚壳贻贝的贝壳开放行为和呼吸代谢[J]. 中国水产科学, 2021, 28(10): 1329-1336.
[2]	潘文澜, 罗定贵. 溶解氧指标的意义和几种测定方法[J]. 当代化工, 2014(7):1397-1399,1402.
[3]	邱小龙, 江颖琳, 蔡雅霜, 等. 雌雄罗氏沼虾应对低氧胁迫的行为生理响应[J]. 应用生态学报, 2022, 33(10): 1-10.
[4]	林国文. 珍珠龙胆石斑鱼对水温、盐度和低溶解氧耐受能力的初步研究[J]. 渔业研究, 2020, 42(5):481-485.
[5]	林欣, 区又君, 温久福, 等. 急性低氧胁迫对四指马鲅幼鱼鳃和肝组织损伤的影响[J]. 渔业研究, 2023, 45(1):14-22. DOI
[6]	徐俊龙, 莫莉莉, 阳连贵. 溶解氧在水产养殖中的重要作用和智能监测[J]. 农村实用技术, 2021(5): 79-80.
[7]	李燕. 水产养殖中的水质问题及解决方法探析[J]. 山西水利科技, 2021(3): 67-69.
[8]	刘双印, 徐龙琴, 李振波, 等. 基于PCA-MCAFA-LSSVM的养殖水质pH值预测模型[J]. 农业机械学报, 2014, 45(5): 239-246.
[9]	袁琦, 黄建清, 符新, 等. 基于神经网络的水产养殖水质预测模型研究[J]. 湖北农业科学, 2013, 52(1): 143-146.
[10]	孙龙清, 吴雨寒, 孙希蓓, 等. 基于IBAS和LSTM网络的池塘水溶解氧含量预测[J]. 农业机械学报, 2021, 52(S1): 252-260.
[11]	张秀菊, 王柳林, 李秀平, 等. 基于BP神经网络的潇河流域水质预测[J]. 水资源与水工程学报, 2021, 32(5): 19-26.
[12]	张学坤. 基于SOA-SVM和SOA-BP模型的溶解氧预测[J]. 人民珠江, 2021, 42(4): 99-104.
[13]	邱章龙. 基于GA-BP神经网络的深基坑变形最优化预测研究[J]. 山西建筑, 2022, 48(19): 143-146.
[14]	Reshef D N, Reshef Y A, Finucane H K, et al. Detecting novel associations in large data sets[J]. Science, 2011, 334(6062): 1518-1524. DOI PMID
[15]	张朝霞, 吴杰. 基于最大信息系数的多变量间相关关系度量方法研究[J]. 太原师范学院学报(自然科学版), 2022, 21(1): 34-40.
[16]	崔莉凤, 黄振芳, 刘载文, 等. 水华暴发叶绿素-a与表征指标溶解氧和pH的关系[J]. 给水排水, 2008, 34(S1):177-178.