| As the source of life,water plays an essential role in human production and life.Almost half of the world’s population relies on water from lakes and reservoirs for drinking,fishing,transportation,tourism,etc.Monitoring in recent years has found that most lakes and reservoirs in China are either heavily or lightly eutrophic.Algal eutrophication is particularly prevalent in lakes and reservoirs in the economically developed eastern region,such as Taihu Lake and Chaohu Lake.Accurate and efficient assessment of eutrophication status is very important for water quality research and management.Among the many indicators for water quality monitoring,Chla concentration can be a good characterization of algal biomass in lakes.Monitoring water quality Chla concentration enables effective and timely assessment of the eutrophication level of the lake,identification of the contaminated areas and the pollution level,and the development of effective treatment measures and actions.However,laboratory assays after field collection of water samples are still the main means of testing for Chla in water quality.The existing online monitoring equipment,with low accuracy of estimation methods and poor field application capability of instruments,can hardly meet the requirements of practical monitoring of water quality Chla.Therefore,the research of accurate,rapid,efficient,in situ,low-cost,and online monitoring technology for water quality Chla is essential for sustainable water ecology development.In this paper,the research of accurate,rapid,efficient,in situ,and lowcost online detection technology of Chla in water quality is carried out based on UVvisible-NIR spectroscopy.The following is the main work and contributions of the paper.1)To address the current water quality Chla monitoring work is still based on manual sampling laboratory analysis and assay,which is challenging to meet the real needs for accurate,rapid,efficient,and low-cost monitoring.This paper experimentally validates the instrument and device parameters and data acquisition methods used in the spectral detection system based on UV-Vis-NIR spectroscopy.The results of the experiments showed that the optimal accuracy of R2 was 0.9287,RMSE was 8.8637 μg/L,and MAE was 7.0639 μg/L for the prediction of chlorophyll a concentration in the sample solution in the laboratory environment.This experiment assesses the feasibility of the design solution and reduces the technical risk of the instrument development.A spectral detection system was constructed using this as a guide.2)To address the problem that many current traditional data processing algorithms perform poorly in practical applications,resulting in low accuracy results.In this paper,we study the spectral feature extraction algorithm in the data processing process in depth.This paper introduces the "Boruta algorithm-based local optimization process "based on the traditional simulated annealing algorithm and proposes the "two-step simulated annealing algorithm(TSSA)." This algorithm combines global optimization and local optimization,and the efficiency of the optimization is greatly improved.The Boruta algorithm ensures that the feature extraction results are all strongly correlated with the dependent variable,reducing data redundancy.The experimental results show that in the data sets processed by first-order differentiation,the accuracy coefficient of determination R2 of the inverse model established by spectral feature extraction using the TSSA algorithm is improved by 24.63%-91.89%;the root mean square error RMSE is decreased by 60.83%-73.62%;the mean absolute error MAE is decreased by 56.92%-70.78% compared with the traditional feature extraction algorithm.3)To address the lack of accurate and efficient quantitative inversion models in spectral water quality testing technology.This paper used the developed spectral detection equipment to take field measurements and collect data in Chaohu and Taihu lakes,typical eutrophic shallow water lakes in China.The absorption spectra were calculated using the spectral data of distilled water as the calibration reference.The spectral features of the data were enhanced using three methods: first-order differentiation,second-order differentiation,and de-envelope.Six methods,such as the correlation coefficient method,band ratio method,and principal component analysis,are used for spectral feature extraction.The extracted features are input to BP neural network for inversion modeling.According to the model evaluation indexes,the inversion model is built using the data processing method of differential spectral calibration → first-order differential processing → Savitzky-Golay filtering → TSSA spectral feature extraction → BP neural network.The accuracy indexes of the inversion model with the determination coefficient R~2 of 0.9654,the root mean square error(RMSE)of 3.6723 μ g/L,and the mean absolute error(MAE)of 3.1461 μg/L.It achieves accurate,fast,efficient,in-situ,and low-cost online acquisition of Chla concentration in water. |
PDF Full Text Request