Estimation of algae growth model parameters by a double layer genetic algorithm

Abstract

This paper presents a double layer genetic algorithm (DLGA) to improve performance of the information-constrained parameter estimations. When a simple genetic algorithm (SGA) fails, a DLGA is applied to the optimization problem in which the initial condition is missing. In this study, a DLGA is specifically designed. The two layers of the SGA serve different purposes. The two optimizations are applied separately but sequentially. The first layer determines the average value of a state variable as its derivative is zero. The knowledge from the first layer is utilized to guide search in the second layer. The second layer uses the obtained average to optimize model parameters. To construct a fitness function for the second layer, the relative derivative function of the average is combined into the fitness function of the ordinary least square problem as a value control. The result shows that the DLGA has better performance. When missing an initial condition, the DLGA provides more consistent numerical values for model parameters. Also, simulation produced by DLGA is more reasonable values than those produced by the SGA.