OPTIMIZATON ALGORITHMS FOR AERATION BASED CONTROL OF WASTEWATER TREATMENT PLANTS

Apply

OPTIMIZATON ALGORITHMS FOR AERATION BASED CONTROL OF WASTEWATER TREATMENT PLANTS

Journal Title: EcoTerra (Journal of Environmental Research and Protection) - Year 2009, Vol 22, Issue 22

Abstract

This paper presents aeration based control strategies using reference data of a benchmark developed for creating and evaluating different control strategies. The aim was to investigate the applicability, reliability and time complexity of two optimization algorithms. These two methods were: the gradient descent (or gradient method), a conventional minimizing algorithm and a stand-alone metaheuristic, the genetic algorithm. The problem, through which the comparison was made, was to give value to the parameters of a modified PID (Proportional, Integral and Derivative) control. The input value was defined by the difference of the dissolved oxygen (DO) concentration set-point (2 g/m3) and the DO concentration of the fifth reactor. The controlled argument was the mass transfer coefficient modified through the change of the air flow rate. The objective was to keep the DO measurements in the interval of 1.8-2.2 g/m3. Optimisation process was carried out with the help of a graphical toolbox. After tuning the parameters manually the constraints were narrowed in multiple steps. Two different sets of parameters were defined as initial values and all cases were run twice. It became clear from the results that the two optimisation techniques gave similar results on the PID parameters, the observed DO curves were very close to each other. The curves with a few exceptions stayed within the range of 1.8 and 2.2 g/m3 despite of the wider optimization interval. The differences between the results indicate that there are more than one local minimiser, which denotes that the gradient descent would not be applicable for the problem without pre-defined boundary conditions. The time complexity of the algorithms was also examined during the simulation. The features of the different methods could be clearly observed. With one exception the gradient method was more time consuming and the time-span varied largely while the time frame of the genetic algorithm was steadily between two and three hours.

Authors and Affiliations

V. SOMOGYI, E. DOMOKOS, Á. RÉDEY

Keywords

EP ID EP97099
DOI -
Views 147
Downloads 0