The Research On Novel Concepts And Algorithms Of Pollutants Biodegradation Kinetics And Quantitative Structure Property Relationship Prediction

In this paper, for the activated sludge kinetics, based on the traditional conception of the influent F/M ratio (fi ) in the activated sludge wastewater treatment technology, some novel terms, such as the effluent F/M ratio (fe), the process decrease of F/M ratio (-f) and the compound half saturation coefficient (Kfs) etc. were introduced. The quantitative relationships between the F/M ratio (including fi, fe and -A/) and main operation parameters of the aeration tank, the dynamic quantitative change of fe along with the time, and the correlations between the F/M ratio and main performance indexes of the aeration tank were therefore set up or discussed. Using real data from a municipal wastewater treatment plant report, the rationality of relevant relationships were validated and it is proved that the fe can be used for measuring wastewater treatment plant operation condition more sensitively than other monitoring indexes. This study can provide a theoretical basis for understanding the meaning and wider application of the food to microorganism ratio.For the microbiology kinetics, the application of properly improved genetic algorithm (IGA) for automatically generating initial parameter estimations for biodegradation kinetic models is described. The suitable starting points provided by IGA are combined with a locally converging nonlinear parameter estimation algorithm, Gauss-Newton method. The proposed hybridization method was employed to estimate the parameters of the classical biodegradation kinetic model, Contois model. Synthetic data and published data are used to verify the presented algorithm. The results show the effectiveness of the algorithm. In addition, a GA with Instantaneous Elitist Protection Strategy (IEPGA), a simply improved genetic algorithm (IGA) and the method including repetitious procedure of lsqnonlin function (Matlab software) combined with initial values generated stochastically within the wide parameters range are respectively employed to search the parameter estimations of the integrated Monod biodegradation kinetic model. The results based on synthetic data and published data show that the proposed three methods are all suitable for this problem. However, in view of the operation scale, runtime and the final results precision, the method including repetitious (>20 times) procedure of Isqnonlin function with stochastic initial value is more convenient than GAs to obtain the biodegradation kinetics parameter.For the enzyme kinetics, a convenient and accurate method for estimating integrated enzyme kinetics parameter is presented using the lsqnonlin function (Matlab software) coupled with weighted nonlinear least-squares analysis. The sum of the squared weighted errors (SSWE) proposed by solving the parameter estimation of Monod model was introduced to solve the similar model, integrated enzyme kinetics model. The weighted nonlinear least-squares analysis can be conveniently employed with all kinds of nonlinear optimization methods. Combined with the powerful search ability of Isqnonlin method, the parameter estimation of integrated enzyme kinetics model can be resolved. The results based on synthetic data demonstrated the validity of the proposed method.For QSAR and QSPR model, Bayesian regulation BP neural network model of quantitative relationship between the electrochemical reduction potential and 8 molecular descriptors of 87 chlorinated aromatic compounds was established. The descriptors consists of 6 quantitative structure descriptors calculated by MOPAC2000 built in ChemOffice2004: homo energies (EHOMO), lumo energies(ELUMO), heat of formation(HF), dipole(DIP) electronic energy(EE), core-core repulsion(CCR), and two ordinary ones: Cl number(Cl) and molecular weight(MW). The optimal network structure achieved is 8-19-1, prior to 5-S-1 (S represents the number of hidden layer node)with the same descriptor as the stepwise linear regression, has the correlation coefficients square and mean square error for the training set and the test set: 0.999 and 0.00011, 0.982 and 0.0015.