Study Of Water Quality Prediction And Evaluation Method Based On PLS-SVM

With the rapid growth of population and development of social economy, waterenvironment is in the progressive deterioration and faced with tremendous protectionpressure, pollution and security problems of water environment quality has become afocus concerns of the whole world. The reasonable evaluation of water quality andforecast objectively valid on water quality factors, is the basis work for waterenvironment management and control, has the important practical significance.Based on the12th Five year national science and technology support program"security type city evaluation index system and evaluation system research anddevelopment" as the research background, the related research on comprehensiveprediction and evaluation of water quality is done. First introduced the present researchsituation and forecast of water quality evaluation at home and abroad, and proposed tothe characteristics of multi correlation among the factors for water quality, Put forwardwater quality prediction and evaluation of the combination model based on partial leastsquares (Partial Least Squares, PLS) and support vector machine (Support VectorMachine, SVM). The results of this study are as follows:The final structure of Support Vector Machine is determined by the kernel functionand parameters,, using the particle swarm optimization algorithm (Particle SwarmOptimization, PSO) for parameter optimization, for the problem that the particle swarmalgorithm is easy to fall into local optimum in the search process, introduced the historyinformation, the particle can move according to the best direction in the search process,at the same time the mutation factor is introducted, enhance the ability to avoid localoptimal random variation. Through the simulation experiment,campared with the gridsearch method, genetic algorithm, particle swarm algorithm are compared, the resultsshow that the improved particle swarm algorithm is more efficient.According to the traditional water quality prediction mainly adopts the method ofsingle factor forecast, forecast model is put forward a comprehensive historicalinformation based on the water quality, and analyzing the factors of water quality datafor the correlation, discovered that there is multiple correlation can brings informationredundancy, declining the forecasting accuracy,bringing in the partial least squaresmethod to realize mathematical reduction of the quality factor, eliminate the redundantinformation, then predicted by means of SVM prediction machine. Through the simulation experiment proves that the accuracy and efficiency of the water qualityprediction model based on PLS-SVM which is proposed in this paper is higher thanother models.In the water quality evaluation process, because of water is a organic interacted byphysical, chemical, biological factors, many pollutants, water quality factors associatedwith diversification, evaluation index, put forward the water quality evaluation modelbased on PLS-SVM. Extracting and compressing information data by partial leastsquares method, mathematical reduction, made new variables do not contain redundantinformation, water quality evaluating by SVM, experimental results show that thismethod has better accuracy and efficiency.