Coagulant Dosage Neuro-fuzzy Control Research And Design

Coagulation is one of the most important processes of water treatment. It can reduce the operating load of filtration and antisepsis equipments by adding coagulant dosing accurately, which is very important to improve water qualities and economic benefits. At present the dosing control of domestic factory is based on streaming current detector (SCD) feed-back control and based on mathematic model feed-forward coagulant dosing control. But the results of both coagulant control schemes are not satisfying. There are some questions of chemicals wastage and sediment output water qualities varieties. It is a pending problem how to calculate the efficient coagulant dosage according currently raw water qualities in water treatment industry .It takes more than 40 minutes to finish coagulant process which includes the projecting of chemicals, mixing, flocculating and subsiding. It is affected by many factors such as many characteristics of raw water quality, capacity of treating water and coagulant technics. Coagulant process is not only a complicity physical-chemical process of difficultly modeling but also a big time-delay, nonlinear and uncertain system. So it is difficult to control by traditional control approaches. The blooming artificial control provides a new way for coagulation control. In this thesis,the water treatment process and the actuality and trend of coagulant dosing control scheme are first introduced. After briefly introduced Neural Network Control (NNC) and Fuzzy Control of IC , Adaptive-Network-Based Fuzzy Inference System (ANFIS) ,one of the combination of NNC and FC, is expatiated in detail. By analyzing the characteristics of coagulant process and the main schemes of coagulant dosing control, NN feed-forward controller and ANFIS feed-forward controller based on raw water turbidity, temperature, PH, alkalinity are designed which can substitute Jar-Test coagulant dosage control. In designing ANFIS scheme, some sample data is classified by subtractive cluster method, and some fuzzy membership functions and rules are obtained, and a initial ANFIS structure is established. NN control model and ANFIS control model are simulated and tested by using Jar-test historical data. Traditional mathematic model is also simulated for comparison. The results of simulation suggest that ANFIS feed-forward control model is distinct superior to the others. It can predict coagulant dosage effectively according to raw water in time. The control performance of NN model is generic, not very good. At last, the scheme of ANFIS feed-forward control of coagulant dosage is discussed on how to practice in engineering.