Inquiry to the Effectiveness of Genetic Algorithms for Accurate Global Optimization of Continuous Functions

For a given precision of optimization, whether and how fast an algorithm achieves it can be used as a testing standard of performance. Compared with the search efficiency of random sampling, the performance of conventional binary coded genetic algorithm (GA) used for accurate global optimization is tested in this paper. It is found that the efficiency of GA in exploration and exploitation is not even better than that of random sampling because of simultaneous operation of these two processes. The Hamming cliff which may occur in the binary coded genetic algorithms is difficult to be bridged by conventional crossover strategies. The accuracy of solution obtained by GA through limited generations of evolution can not be well assured and evaluated. Based on the results, the strategies to improve the performance of GA are suggested.