Thesis (MTech (Electrical Engineering))--Cape Peninsula University of Technology, 2007 / In order to develop an effective control strategy for the activated sludge process of a
wastewater treatment plant, an understanding of the nature of the influent load disturbances
to the wastewater treatment plant is necessary. Biological systems are among the most
difficult to control and predict. Due to the complex biological reaction mechanisms, the highly
time-varying, and multivariable aspects of the wastewater treatment plant ('MNTP), the
diagnosis of the 'MNTP are still difficult in practice. The application of intelligent techniques,
which can analyse the multi-dimensional nonlinear process data using a visualisation
technique, can be useful for analysing and diagnosing the activated-sludge process in the
VVVVTP. This complex capability for nonlinearity representation combined with the fact that no
model exists for the WVVTP influent dynamics to a WVVTP, makes neural networks an ideal
choice for a solution.
Forecasting the behaviour of complex systems has been a broad application area for neural
networks. Applications such as economic forecasting, electricity load I demand forecasting,
and forecasting natural and physical phenomena have been extensively studied, hence the
numerous papers presented at annual conferences in this focus area. The cognitive ability of
artificial neural networks to map' nonlinear complex input-output relationships, which would
allow for better prediction and corrective control of processes, make them particularly
attractive.
The values of the influent disturbances are usually measured off-line in ~ laboratory, as there
are still no reliable on-line sensors available. This work presents the development of a neural
network model for prediction of the values of the influent disturbances based on historical
plant and weather data, which ultimately affect the activated sludge process. Three different
neural networks including the multilayer perceptron, recurrent and radial basis functions
neural network are developed for the prediction of the influent disturbances of Chemical
Oxygen Demand (COD), Total Kjeldahl Nitrogen (TKN) and influent flow rate respectively.
The application area is the prediction of the influent variables at a local municipal wastewater
treatment plant. The forecast result is used for the determination of the setpoint to a
controller, in order to• optimize plant performance. The results are first applied to a pilot
wastewater treatment plant.
Much hype exists surrounding the subject of neural networks, and they are sometimes
described as 'computers that think'. This sort of definition creates unrealistic expectations
and is one of the reasons why it is discredited. The results obtained will hopefully present
helpful insights as to the scope and possibilities as to the application for neural networks, but
also present the practical challenges which neural network practitioners and designers of
intelligent systems face.
The solution of the problem for development of the mathematical model for dynamic
behaviour of the influent disturbances according to the influence of the weather conditions
and the season of the year is the first attempt in the scientific and research literature so far.
Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:cput/oai:localhost:20.500.11838/1112 |
Date | January 2007 |
Creators | Kriger, Carl |
Publisher | Cape Peninsula University of Technology |
Source Sets | South African National ETD Portal |
Language | English |
Detected Language | English |
Type | Thesis |
Rights | http://creativecommons.org/licenses/by-nc-sa/3.0/za/ |
Page generated in 0.0014 seconds