Research On Multi-Parameter Water Quality Detection Method Of Seawater Based On UV-VIS Spectrum

China is a maritime resource-rich country,but with the development of the ocean economy,seawater pollution is increasing.The traditional method for seawater quality testing is based on chemical detection,which not only produces secondary pollution but also requires manual collection of water samples and laboratory analysis,making the process time-consuming and laborious.Therefore,there is an urgent need for a convenient,fast,real-time,and remote seawater quality testing method to solve the problems of the tedious and time-consuming steps of the chemical method.Currently,seawater quality testing involves multiple parameters,with the most critical being chemical oxygen demand,total nitrogen,ammonia nitrogen,total phosphorus,and turbidity.Therefore,this paper studies the multi-parameter seawater quality testing based on UV-visible spectroscopy.By establishing a nonlinear mapping relationship between the multi-parameter feature spectra and the concentrations of various parameters to be measured,the prediction of multiparameter concentrations in seawater and the monitoring of multi-parameter water quality can be achieved.Firstly,ultraviolet-visible spectra of seawater with different concentrations of multiple parameters were obtained through experiments using LS-3000 high-power halogen light source,GZ-tex fiber optic spectrometer,glass cuvette and CZspviewer V2.11 software on a computer.Secondly,due to the large amount of redundant information and spectral noise in the spectra data obtained from the experiments,a competitive adaptive re-weighted sampling algorithm was applied for feature extraction,which retained more feature spectral data and complete material spectral features.Thirdly,to establish a seawater multi-parameter water quality detection model,three multi-input and multi-output neural network structures were constructed based on different neural network principles.Furthermore,a distance-based multiobjective particle swarm optimization algorithm was used to perform multi-objective optimization on the BP neural network.Finally,the feature spectral data extracted by CARS was inputted into the neural network for training,resulting in three seawater multi-parameter water quality detection models(CARS-LSTM,CARS-CNN,CARS-DISMOPSO-BPNN).The experimental UV-visible spectra of the seawater multi-parameter mixed solution and the mixed UV-visible spectra of each measured substance have deviations.After compensating for the errors,the characteristic spectral data are input into the three seawater multi-parameter water quality detection models.The determination coefficient of the CARS-DISMOPSOBPNN model is higher than that of the CARS-LSTM model by 5.61% and higher than that of the CARS-CNN model by 0.75%.The average relative error,average absolute error,and root mean square error of the CARS-DISMOPSO-BPNN model are much smaller than those of the CARS-LSTM and CARS-CNN models.Therefore,the proposed CARS-DISMOPSO-BPNN seawater multi-parameter water quality detection model has higher detection accuracy and the smallest residuals between predicted and true values.The proposed method for seawater multiparameter water quality detection is simple to operate,fast in detection speed,has no secondary pollution,can be remotely detected,and can overcome the interference of spectral distortion of mixed solutions during detection,ensuring that chemical oxygen demand,total nitrogen,ammonia nitrogen,total phosphorus,and turbidity are accurately detected as multiple targets with the best effect.