For the issue of poor robustness of genetic algorithm (GA) in optimization o f optical systems with super-multiple parameters,an improved genetic algorithm is developed by incorp orating normalization of parameters and self-adaptive mutation probability into the real-coded genetic algorithm mixed with escape function (MERCGA).The discussed algorithm (ANRCGA) and MERCGA are applied to o ptimize a fisheye lens whose imaging performances resulting from the two algorithms are ev aluated by ray-tracing calculation with Zemax and evaluation function for comparison.The study shows t hat the former has better imaging performance than the latter as well as the reference design from the pa tent.The proposed algorithm has better robustness and calculation efficiency in optimization of optical syst ems.