Design of Internal Model Control for Wastewater Treatment Plant Using Artificial Neural Network: Case Study Kality Wastewater Treatment Plant
No Thumbnail Available
Date
2021-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
High non-linearity of wastewater treatment processes and the influent wastewater dynamics dependency on the season and surrounding lifestyle, it is still an active research area that requires a new control methodology to optimize and improve the current controller performance. Hence, this research is designed to develop an artificial neural network plant and controller models for the Kality wastewater treatment plant using feedforward backpropagation algorithms. Using the Levenberg Marquard learning algorithm, the plant model was developed to estimate the dissolved oxygen in the biological reactor with six input variables. After several ANN architectures are tested, the best result is obtained with three layers of 43 neurons each, a sigmoid activation function on the hidden layers, and a linear activation function in the output node. Similarly, an inverse ANN (internal model controller) model is designed to control the oxygen transfer coefficient with six input variables. The final optimal inverse neural network model structure is three layers with 47 neurons per layer, a sigmoid activation function on the hidden layers, and a linear activation function in the output node with the Levenberg Marquard learning algorithm. The simulation results also demonstrated that the performance of an ANN internal model controller outperforms that of a traditional PI controller in both transient and steady-state conditions. In the case of the PI controller, the plant response takes 19.4 seconds to reach a steady-state, whereas the ANN internal model controller requires only 16 seconds. This result shows that the ANN controller improves settling time by 21.25 percent. When the rise time of the plant response is evaluated, the ANN internal model controller improves the rise time by 67% compared with the PI controller. Since, ANN plant and controller models developed using process state variables, which provide a significant opportunity for disturbance rejection, transient and steady-state performance improvement.
Description
Keywords
Artificial Neural Network, Feedforward Backpropagation Algorithm, Internal Model Control, Levenberg-Marquardt Algorithm, PI, Wastewater Treatment Plant