Application Of Multi-kernel SVR In Modeling Indicators Of Water For Sewage Treatment

Sewage treatment plants have made a significant contribution to water protection as well as waste water prevention.Since sewage treatment system which is a big lagging, complex, non-liner,strong-coupling and multivariate. Its mechanism study is immature.Coming results are that real-time measurement of key parameters cannot be realized while the sewage treatment effectiveness is highly dependent on the quality of the output water. Thus it is of practical significance and application value to establish a reasonable sewage effluent index model which is highly efficient and reasonable to guide the running of sewage treatment plants, and adjust running condition of different procedures during the treatment process dynamically.Aimed to deal with such problems as rather complex distribution of water quality parameters during the sewage treatment process and inaccurate modeling of single-core support vector regression machine,this paper,on the basis of previous research achievements and taking sewage sludge quality after the using of activated sludge process as the studying object, combining multi-kernel and intelligent algorithm to establish multi-kernel support vector regression(MK-SVR) model of water quality parameters discharged. The studying points are as follows.First,after an understanding of the sewage treatment procedures as well as methods and an analysis of water parameters as well as sewage discharge standard affecting sewage treatment procedures, principal component analysis(PCA) is adopted to deal with dimension reduction of such affecting factors and a new principal component is used to input SVR so as to establish MK-SVR model of effluent water chemical oxygen demand(COD), biochemical oxygen demand(BOD), suspended solids(SS) and total nitrogen(TN).Then,because of the problems this model itself contains and the characteristics particle swarm optimization(PSO) has(easy programming, simple structure, easy to complement, fast searching ability, strong convergence ability), the author puts forward an idea to use PSO to optimize parameters of MK-SVR model and further perfect PSO in terms of itsshortcomings.Finally, to make this model suggested more persuasive,the author makes a comparison of the forecast results of these different models, namely single-kernel SVR, multi-kernel SVR, single kernel SVR and multi-kernel SVR on PCA basis,PSO+MK-SVR and after the perfection of PSO+MK-SVR on PCA basis. Analyzes several performance indexes such as relative error, mean-square error and correlation coefficient. As the results shows, based on PCA analysis of improved PSO+MK-SVR is of the most effectiveness and strongest generalization performance and can be a powerful theoretical support of the highly efficient and real-time running of sewage treatment plant.