基于遗传算法优化BP神经网络的青藏高原海北高寒湿地CO_(2)通量模拟及其影响因子  

Simulation of alpine wetlands CO_(2) flux and its influencing factors based on BP neural network optimized by genetic algorithm in Qinghai-Tibet Plateau

作　　者：曹镓玺 王鑫 雷光春[1] CAO Jia-xi;WANG Xin;LEI Guang-chun(School of Ecology and Nature Conservation,Beijing Forestry University,Beijing 100083,China;Institute of Remote Sensing Science and Engineering,Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China)

机构地区：[1]北京林业大学生态与自然保护学院,北京100083 [2]北京师范大学地理科学学部遥感科学与工程研究院,北京100875

出　　处：《山东大学学报（理学版）》2021年第5期33-41,50,共10页Journal of Shandong University(Natural Science)

摘　　要：构建了使用潜热通量、显热通量、空气温度、总辐射、有效辐射、土壤温度、土壤体积含水量来模拟湿地生态系统CO_(2)排放通量的3层BP神经网络。在确定BP神经网络拓扑结构之后,使用遗传算法取得了BP神经网络训练的最优初始阈值和权值。采用2190组青藏高原海北高寒湿地生态系统中实测CO_(2)通量及相应的训练数据归一化处理后,对所建立的BP神经网络进行训练和验证。验证结果显示,训练组、验证组和测试组的回归系数分别为0.942、0.935和0.938,总体回归系数为0.941,模拟值与实测值之间的均方根误差为0.92%,高寒湿地生态系统中CO_(2)模拟通量为1.207~13.767 g/(m^(2)·d),均值为6.008 g/(m^(2)·d)。相关性分析结果显示,青藏高原海北高寒湿地生态系统中CO_(2)通量与输入神经元的相关系数r表现为:土壤温度(0.77)>空气温度(0.72)>土壤体积含水量(0.39)>潜热通量(0.30)>显热通量(0.29)>总辐射(0.09)>有效辐射(-0.02)(P<0.01)。青藏高原海北高寒湿地生态系统中CO_(2)通量主导因子为土壤温度和空气温度,后续建模工作中应提高这二者的权重。A three-layer BP neural network was constructed to simulate CO_(2) emission flux of wetlands ecosystem by using latent heat flux,sensible heat flux,air temperature,total radiation,effective radiation,soil temperature and soil volume water content.After determining the topological structure of BP neural network,the optimal initial threshold and weight of BP neural network training are obtained by genetic algorithm.The established BP neural network is trained and verified by normalizing 2190 sets of measured CO_(2) fluxes and corresponding training data in wetlands ecosystem in Haibei alpine wetlands of Qinghai-Tibet Plateau.The validation results show that the regression coefficients of training group,validation group and test group are 0.942,0.935 and 0.938,respectively.The overall regression coefficient is 0.941,the root mean square error between simulated and measured values is 0.92%,and the simulated CO_(2) flux in wetlands ecosystem is 1.207-13.767 g/(m^(2)·d),with an average value of 6.008 g/(m^(2)·d).Correlation analysis shows that the correlation coefficient r between CO_(2) flux and input neurons in alpine wetlands ecosystem is:soil temperature(0.77)>air temperature(0.72)>soil volume water content(0.39)>latent heat flux(0.30)>sensible heat flux(0.29)>total radiation(0.09)>effective radiation(0.02)(P<0.01).The dominant factors of CO_(2) flux in alpine wetlands ecosystem are soil temperature and air temperature,and the weights of these two factors should be increased in the subsequent modeling work.

关 键 词：CO_(2)排放通量 Matlab 遗传算法 BP神经网络 湿地生态系统 

分 类 号：X171[环境科学与工程—环境科学]