Intelligent Modeling And Optimization Control Technique Application Research For Biochemical Wastewater Treatment

Activated sludge wastewater treatment is a method by simulating water self-sanitization process which utilizes microorganic life activity to purge organic pollutants in wastewater. It is the main way to the world's industrial organic wastewater treatment and secondary treatment of urban sewage.Becaues wastewater treatment process has the character of nonlinear, time-varible, uncertain and hysteresis and control process is very complex, the traditional control methods are difficult to carry out real-time control.Recently, the introduction of artificial intelligence (AI) is a strong impetus for research on control technology about wastewater biochemical treatment process. But while existing research has made some achievements, research has emphasis on fuzzy control, neural networks, genetic algorithms, expert systems and so on. This paper focuses on the characteristic of urban sewage treatment to research intelligent modeling and optimization control method based on LS-SVM.This thesis comes from National Natural Science Foundation of China (No.60774032), Special Research Fund of Ministry of Education of China for College Doctoral Subject (Project for Young Scholar) (No.20070561006) and GuangDong Province Natural Science Foundation for doctor to start project (No. 9451064101002853), and the research object is the wastewater treatment technique in GuangZhou LiJiao Wastewater Treatment Plant. Utilizing the AI technologies such as ARMA model, Kalman filtering, data fusion, artificial immunity and LS-SVM, some modeling and control problems in the biochemical treatment process, including the abnormal data detecting, Kalman filtering measurement fusion, output water indices prediction model are in-depth studied. Results with practical and theoretical significance have been obtained.The main results and innovations of this dissertation are as follows.1. Aiming at the problem that measurement machines sometimes can detect abnormal data, the reality/anomaly detecting mechanism and negative selection with 3σprinciple algorithm is proposed to check up the abnormal value, and the reality set and antibody set which represent the character of the normal data and the abnormal data each other are built up. In this way, the abnormal data can be detected by data sets or direct method. The simulation results show the method is suitable to check up wastewater data and can effectively monitor operational status of measuring instruments.2. AR model is proposed to amend the abnormal data. AR model is a time series analysis method; it is an algorithm which uses the historical information and the reciprocity of historical information to predict the future trail. The rank of AR model is decided according to information theory rule, and then the AR model is built up by Burg algorithm. When the abnormal value is detected, it can be corrected by historic normal value. The algorithm amending the abnormal value has higher precision, good stability and is suitable to modify the data about wastewater treatment process.3. Considering the problem about multisensor data in the process of wastewater treatment, a multisensor data fusion algorithm is proposed based on the correlative function and least square. It means that the data of sensors with high support degree are fused and data of low level support degree sensors don't be fused. The algorithm has simple calculation and has high fusion precision because the prior information is not necessary. And then an advanced least square algorithm is presented and the advanced fusion algorithm is better than the non-advanced fusion algorithm.4. The calculated value of reaction equation about wastewater treatment process is taken as reference value, the Kalman filtering measurement fusion algorithm is presented based on non-linear system, including a new measurement fusion algorithm based on ridge estimation. On the basis of risk function, it is proved that the new algorithm is superior to other algorithms. The simulation results show the superiority of the Kalman filtering measurement fusion algorithm.5. A prediction algorithm applied to output water indices of wastewater treatment process is proposed. Fisrtly, an adaptive self-recurrent wavelet neural network with chaotic dynamic value is presented. And then the predictive model is set up based on LS-SVM and a new hybrid mutation immune algorithm is proposed to optimize the LS-SVM model parameters. By using stochastic processes martingale theory, it is proved that the algorithm is strong convergent almost everywhere and even has a stronger convergence. The simulation results show that predictive model has high accuracy of prediction.6. A kind of algorithm for multivariable optimal control of waste water treatment process with the lowest operating costs by multi-constraint conditions is presented. Output water indices can be predictive using LS-SVM prediction model. On the condition of satisfying output water requirement of COD, SS and NH3-H, a mathematic model of waste water treatment proeess is developed for multivariable optimal control problem. The simulation results show that the method can reduce production costs.