| The artificial neural network is an emerging artificial intelligence modeling technique which has a great potential in environmental engineering and especially related to treatment processes. In this thesis, an artificial neural network modeling approach was used for two applications in water treatment. First, an artificial neural network model was built to forecast the raw water colour in the North Saskatchewan River. Similar models for the other raw water quality parameters can be established by using the approach outlined in the thesis. This thesis proceeds to provide a survey of recent research applications of neural networks in two process control strategies: adaptive control and internal model control. From the analysis of the benefits and deficiencies of these two control strategies, and the associated engineering knowledge of the water treatment process, a feedforward neural network controller was proposed and built for the Rossdale water treatment plant in Edmonton, Alberta. Upon the success of both models, this thesis also attempts to explain why neural network modeling functions well from the viewpoints of inductive learning and the computational theory. The methodology of making the hypothesis space tractable in order to obtain the optimum neural network model was also demonstrated. |
