Modeling of the chemical dosing at a water resource recovery facility (WRRF)

Abstract

The focus here in this master thesis is developing a smart artificial neural network to model wastewater dosing. In order to use this model in the optimization process to optimize using chemical dosing materials. The thesis gives the necessary introduction to modeling dynamic systems, machine learning, and neural network. Then the dynamic system is modeled in a classical way, as first-order and second-order transfer function without and with time delay. Afterward, time delay estimation is developed and tested with the simulator.

Machine learning and artificial intelligence neural network (ANN) of the dynamic system is derived from the mathematical model of the dynamic system. The ANN is implemented by using the data from the simulator. The ANN result is compared with the result from the simulator and shows a high performance of the ANN model.

The ANN is implemented in a real process in different methods and software. The implementation has been performed as a multi-input single-output system and multi-input multi outputs system. The difference between methods is discussed.

NARX neural network model gives high performance and good accuracy in a dynamic system. It can deal with time series and handle time delays automatically.