Analysis of Nitrate in Seawater of Wheat Island Based on LLE-BPNN
WANG Xue-ji1, 2, HU Bing-liang1*, YU Tao1, LIU Qing-song1, 2, LI Hong-bo1, 2, FAN Yao1
1. Laboratory of Spectral Imaging Technique, Xi’an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences, Xi’an 710119, China
2. University of Chinese Academy of Sciences, Beijing 100049, China
Abstract:Excessive nitrate in water may influence some aquatic organisms’ survival and cause harm to humans, especially infants. Therefore, nitrate concentration becomes an important indicator in water quality monitoring. Due to the complexity of operation and slow response of conventional methods for measuring nitrate concentration, many researchers have begun to use ultraviolet/visible (UV-Vis) spectroscopy combined with artificial neural network (ANN) methods to measure nitrate content in water. This paper proposes a modeling method combining locally linear embedding (LLE) in manifold learning with back propagation neural network (BPNN). The relationship between the spectral curve of nitrate and the concentration was obtained, so that a rapid and accurate quantitative analysis of the nitrate concentration in the wheat island of Laoshan District, Qingdao was achieved. In the experiment, we selected 59 groups of spiked solutions with different concentrations of filtered wheat island seawater, and collected spectral measurements of these samples using a laboratory-developed spectrum analyzer, with standard normal variate (SNV) method calibrating spectral data of measured nitrate solution to reduce the noise caused by the instrument itself or the environment. First 1 500-dimensional of the pre-processed spectral data was used to avoid insufficient memory when using the entire 2 048-dimensional data to build BPNN model, and a control experiment was performed. Then the number of neighboring points k and the embedding dimension d in the LLE were optimized by the grid search combined with the ten-fold cross validation method, obtaining the optimal k=15, d=3. Then the dimension of the experimental data was reduced. The spectral information of the reduced-dimensional training set and its corresponding concentration information were modeled by the BPNN to achieve a quantitative analysis of the nitrate concentration in the prediction set. Coefficient of determination (R2) and root mean square error of prediction (RMSEP) were introduced to evaluate modeling effects. And compared with the predicted results obtained by only using BPNN modeling, R2 of our improved method increased from 0.926 3 to 0.992 8, and RMSEP decreased from 0.442 5 to 0.280 4, and prediction modeling program run time decreased from 327 s to about 0.5 s. In addition, we used all 2 048 dimensions of the 59 data sets for LLE-BPNN modeling, with R2=0.995 7 and RMSEP=0.136 5, which was improved compared to the modeling accuracy when only using the first 1 500 dimensions, while elapsed time was similar. The analysis results above showed that using the LLE-BPNN method can achieve a rapid prediction of nitrate concentration in seawater, while significantly improving prediction accuracy and reducing prediction time.
Key words:Nitrate concentration; Ultraviolet/visible spectral technology; Locally linear embedding; Back propagation neural network
王雪霁,胡炳樑,于 涛,刘青松,李洪波,范 尧. 基于LLE-BPNN的小麦岛海水硝酸盐含量分析[J]. 光谱学与光谱分析, 2019, 39(05): 1503-1508.
WANG Xue-ji, HU Bing-liang, YU Tao, LIU Qing-song, LI Hong-bo, FAN Yao. Analysis of Nitrate in Seawater of Wheat Island Based on LLE-BPNN. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2019, 39(05): 1503-1508.
[1] ZHANG Yi-wen, LUO Jian-zhong, CHEN Yu-yang(张懿文,罗建中,陈宇阳). Guangdong Chemical Industry(广东化工), 2015, 42(14): 99.
[2] Hu Yingtian, Wen Yizhang, Wang Xiaoping. Sensors and Actuators, B: Chemical, 2016,227: 393.
[3] Hao Z, Huang X, Zhu C. Proc. Int. Conf. Nat. Comput., 2008, (3): 394.
[4] Siddiquee M, Hossain M. Neural Computing and Applications, 2015, 26(8): 1979.
[5] Meilin W, Youshao W, Jidong G. Ecotoxicology, 2015, 24(8): 1632.
[6] Ostadaliakari K, Shayannejed M, Chorbanizadehkharazi H. Ksee Journal of Civil Engineering, 2016, 21(1): 1.
[7] BAO Cui-mei, HAN Xiao-chun, YI Hui, et al(薄翠梅,韩晓春,易 辉,等). CIESC Journal(化工学报),2016,67(3):925.
[8] Chen Y, An S, Dong J. et al. Optical and Quantum Electronics, 2016, 48(11): 488.
[9] Yang T M, Fan S K, Fan C, et al. Environmental Monitoring and Assessment, 2014, 186(8): 4925.
[10] ZHOU Zhi-hua(周志华). Machine Learning(机器学习), 2016.
[11] Fan Dayong, Yang Jiachen, Zhang Junbao, et al. IEEE Journal of Translational Engineering in Health & Medicine, 2017, PP(99): 1.
[12] HOU Ya-li, LI Tie(侯亚丽,李 铁). Journal of Detection & Control(探测与控制学报), 2008, 30(1): 53.
[13] Yen C T, Huang Y J. Multimedia Tools and Applications, 2015, 75(16): 1.
[14] Jiang M, Spikes K T. Geophysical Journal International, 2013, 195(1): 315.