Research On Application Of The Recurrent Support Vector Regression And The ASM2D Model In The Wastewater Treatment Plant Design

To predict the influent and effluent qualities accurately, grasping characteristics of wastewater quality were the premise to design wastewater treatment structures and optimize operation management. The iteration algorithm of support vector machine and the ASM2D model were used respectively to predict the influent and effluent qualities, relying on the practical design of a municipal wastewater treatment plant. And the results verified that the design of wastewater treatment based on the influent qualities predicted by RAVR would help to improve the rationality and economy of design. The main conclusions were as follows:(1) Anew water quality prediction model, the gray support vector regression (GSVR) was proposed to predict the influent qualities of some municipal wastewater treatment plant. Then the accuracy and effectiveness of the model were verified through the acrual of influent qualities. The results indicated that the accuracy of this model is higher than single gray theory model, support vector regression model and BP neural network model, but it has some defects to the prediction of the fluctuation data.(2) On the basis of further improvement of gray support vector regression (GSVR), the recurrent support vector regression machine(RSVR) was proposed which is the innovative point in this thesis. In the actual simulation application, this kind of algorithm showed the higher fitting precision, especially to the prediction of the fluctuation data, which was suitable to be applied in the operation management of the wastewater treatment plant.(3) For the application of ASM2D was based on the COD components measurement. On the basis of the research, this thesis concluded a suite of COD components detection methods which are easy, practical, complete and available to promote and apply in the municipal wastewater treatment plants. All the measures could be finished in one day.①The determination of solubility and non-degradable COD component (SI)the feasibility of physicochemical method to detect SI was verified through the comparison of separation test. Considering from accuracy, practicality and rationality, the method of "flocculation and0.45μm filter membrane" was used as the separation method.;②The determination of rapid and readily biodegradation COD component (Ss)—the feasibility of batch OUR method was verified by tests;③The determination of slowly biodegradation COD component (Xs) was close to BCOD value, which could be determined by BOD5or the mean of BOD5and KBOD:④5-point titration was easy and time-saving due to the specialized data analysis program-TITRA5.EXE and it was recommended to detect the fermentation products (SA). Compared with the SA measurement in real wastewater by HPLC, acetic acid standard solution was measured by5-point to verify the accuracy of the5-point pH titration. The results indicated that this technology is feasible when the acetic acid concentration was in10-60mg/L.(4) The COD components detection methods was applied in the ASM2D model to simulate the effluent qualities.The results showed that biological phosphorus removal process need add chemical phosphorus removal in order to ensure the water quality standards.(5) The actual case illustrated the value of the application of the Recurrent Support Vector Regression to the prediction of input water quality, and ASM2D model to the prediction of output water quality.