Floating genetic algorithms (FGAs) are optimization methods simulating the natural evolution mechanism.FGAs have been widely used in science and technology by virtue of their simplicity

robustness

freedom of calculating the gradient of the objective function

high precision and the ability of solving multi-dimensional numerical problems.With the convergence analysis of FGAs

it is proved in this paper that FGAs with the fittest individual holding in each generation can converge to the global optimum while simple FGAs can not.In the light of the theoretical convergence analysis

improved FGAs with the fittest individual holding and the continuous mutation are proposed

which overcome the incompatibility between the high precision and low computational cost.The improved FGAs have been applied to extracting the semiconductor device model parameters

and have gained about 27% reduction to the computational cost.