Modelling of Activated Sludge Wastewater Treatment

Activated sludge wastewater treatment is a highly complex physical, chemical and biological process, and variations in wastewater flow rate and its composition, combined with time-varying reactions in a mixed culture of microorganisms, make this process non-linear and unsteady. The efficiency of the process is established by measuring the quantities that indicate quality of the treated wastewater, but they can only be determined at the end of the process, which is when the water has already been processed and is at the outlet of the plant and released into the environment.If the water quality is not acceptable, it is already too late for its improvement, which indicates the need for a feed forward process control based on a mathematical model. Since there is no possibility of retracing the process steps back, all the mistakes in the control of the process could induce an ecological disaster of a smaller or bigger extent. Therefore, models that describe this process well may be used as a basis for monitoring and optimal control of the process development. This work analyzes the process of biological treatment of wastewater in the Velika Gorica plant. Two empirical models for the description of the process were established, multiple linear regression model (MLR) with 16 predictor variables and piecewise linear regression model (PLR) with 17 predictor variables. These models were developed with the aim to predict COD value of the effluent wastewater at the outlet, after treatment. The development of the models is based on the statistical analysis of experimental data, which are used to determine the relations among individual variables. In this work are applied linear models based on multiple linear regression (MLR) and partial least squares (PLR) methods. The used data were obtained by everyday measurements of the quantities that indicate the quality of the input and output water, working conditions of the plant and the quality of the activated sludge. The database contains 223 groups, each with 26 parameters, for the entire year of 2004. The variables were analyzed for determination of the most important factors. The analyses were done with significance level of p KPK= 16 mg L-1-O2and the average error of piecewise linear regression model is γKPK= 16 mg L-1- O2