Research And Software Design Of Multi-parameter Water Quality Detection Algorithm Based On Spectrum Method

Water is an essential resource for the production and development of human society.At present,the abuse of water resources and the water pollution occur frequently in China,which do great harm to our life and health.Water pollution has become a thorny issue in the world.Water quality monitoring is an important means to prevent water pollution.The commonly used water quality detection methods include chemical analysis and spectral analysis.The former is mainly based on field sampling and laboratory analysis.It has the disadvantages of tedious operation,requiring pretreatment of chemical reagents,high detection cost,long analysis cycle,easy to cause secondary pollution and so on.It is difficult to adapt to the requirements of environmental monitoring.The new spectral analysis method has the advantages of fast test speed,high accuracy,good repeatability,no reagent consumption,no secondary pollution and can measure multi-parameter simultaneously.Therefore,it becomes the development trend of water environment dynamic monitoring.Compared with foreign advanced technology,there is still a big gap in domestic development.Water quality monitoring equipment on the market is mostly foreign,with only a few parameters of the online detection capacity.With the increase of open parameters,the cost is further increased.Therefore,improving domestic technology and equipment capacitty in water monitoring field is the development direction in future.The parameter quantitative analysis model is a key factor affecting the accuracy of water component prediction.In the past research,the quantitative analysis methods mostly consider the use of multiple linear regression,partial least square(PLS)and other linear regression prediction methods.Because the actual water components are more complex and diversified,the water parameters and absorbance are not linearly correlated,so it is difficult to obtain satisfactory prediction results by using linear prediction model.In recent years,there are also artificial neural network(ANN),support vector machine(SVM)and other non-linear machine learning prediction methods.Therefore,how to combine the advantages of machine learning algorithm and new spectral method in water quality monitoring is a valuable research direction.In view of this,this paper carried out research on multi-parameter water quality monitoring technology based on the spectrum method.The main research contents are as follows:(1)Combining the spectroscopic method with the electrode method,the double opticle path active correction method is used to measure p H,conductivity,temperature,residual chlorine,turbidity,chemical oxygen demand(COD),nitrate nitrogen and total organic carbon(TOC),chroma and UV254 at the same time in a wide spectral range,which concentrate the detection ability of multiple independent sensors on the integrated cabinet,realize the demand of multi parameter real-time,online,fast and dynamic monitoring.(2)The upper computer software of multi-parameter water components on-line monitoring system is designed and implemented.The system design and detailed design of the software are described in detail,including data input,parameter concentration calculation,interface design,and concentration calculation result transmission.By combining GPRS communication technology with modern network technology,Water quality parameters can be monitored in real time on PC and mobile terminals,and the change of water components can be viewed dynamically.(3)To solve the problem of low efficiencecy of the existing algorithm,an support vector regression optimized improved grid search(GS-SVR)is proposed,and a quantitative analysis and prediction model of nitrate nitrogen and chroma is established.The solution sample test experiment is carried out and compared with the current popular algorithms,including back propagation neural network(BPNN),SVR,GS-SVR,particle swarm optimization(PSO)algorithm to optimize SVR and genetic algorithm to optimize SVR.The experimental results show that the optimal parameters C andσare(512,0.0442),R~2=0.9935,RMSEP=0.0435,and the average training time is 13s.Compared with BPNN,SVR,GS-SVR,PSO-SVR and GA-SVR,R~2 increased by 1.22%,11.66%,0.78%,0.74%,0.77%,and training efficiency increased by 4.15 times(BPNN),8.30 times(GS-SVR),21.38 times(PSO-SVR)and10.23 times(GA-SVR)respectively.For the chroma,R~2=0.9840,increased by 1.43%,3.11%,1.02%,0.13%,0.05%respectively.The average training time was 11s,and the efficiency increased by 4.82 times(BPNN),8 times(GS-SVR),23.27 times(PSO-SVR)and 10.55 times(GA-SVR)respectively.It not only improves the prediction accuracy of the model,but also improves the efficiency of optimization.