Research And Implement Of Soft-Computing Model For Total Phosphorus Based On Recurrent Fuzzy Neural Network

During the recent decades, the increased awareness about the negative impact of eutrophication in the quality of water bodies have given rise to more stringent wastewater treatment requirements and regulations. Phosphorus is one of the main factors that cause the eutrophication phenomenon in water system, and it could lead to an abnormal reproduction of aquatic organisms. To solve this problem, we have to restrict the discharge quantity of phosphorus in wastewater treatment process. Therefore, measuring effluent total phosphorus(ETP) online is vital to control ETP concentration in wastewater treatment processes(WWTPs), and it is an important way to improve the regulation level of WWTPs.However, due to the time-delay responses of ETP associated with the transportation of water samples and the complicated chemical mechanism of ETP, the laboratory measurements cannot meet the online monitoring regulations nowadays. Moreover, since the complex biological characteristics of the activated sludge process, building an accretive mathematical model of ETP is almost impossible. Therefore, to achieve the real-time detection of ETP, this study is proposing to combine the partial least square(PLS) method and the recurrent fuzzy neural network(RFNN) to develop a dataderived soft-sensor for online monitoring of ETP. And a monitoring system, using this data-derived soft-sensor, is designed and tested in real WWTPs.The main research results in this paper are listed as follow:(1) Selected variables that relative to ETP. Based on the biological characteristics and history data of WWTPs, PLS algorithm was used in this paper to select variables that have strong relationship with the changing trend of ETP. In this paper, 5 process variables, including temperature, pH, ORP in anaerobic part, DO in the first tank of oxic part and TSS in the last tank of oxic part, out of 11 were selected as the inputs of the soft-sensor model of ETP.(2) Developed the soft-sensor model of ETP. To measure ETP online, RFNN was employed to extract the relation between the selected variables and ETP. Besides, history data were used to train and test the RFNN model. Experiment results have showed that RFNN model of ETP could achieve better prediction results than standard fuzzy neural network(FNN) and other model building technologies.(3) Designed and developed the intellectual soft-sensor system of ETP. At first, a hardware system was built to obtain the data of selected variables in real-time. Then, encapsulated data collection module, data transmission part, soft-sensor part and online measurement part into a system. Finally, using C# language to develop the UI part to display the online detecting results of ETP on monitor device. Moreover, the intellectual soft-sensor system of ETP was tested in real WWTP and achieved a satisfactory measuring result.