Shape reconstruction of a perfectly conducting cylinder using real-coded genetic algorithm

Electromagnetic inverse scattering has been studied extensively in previous years. Most of the previous inversion algorithms an gradient based. They converge rapidly but they are prone to converge to a local optimum. A good initial guess is required to obtain a convergent solution. Genetic algorithms (GAs) are a set of slowly converging probabilistic global optimization methods based on genetic recombination and evolution in nature. They are less prone to converge to a local optimum even when the initial guess is far away from the exact one. The GAs have been widely used for solving electromagnetic problems. Chiu and Liu (see IEE Proc. Pt.H, vol.143, no.3, p.249-59, 1996) used the simple genetic algorithm (SGA) to reconstruct the image of a perfectly conducting cylinder. However, some significant errors are found in the method of Chiu et al. that makes the reported results questionable. In addition, as far as we know, the real-coded genetic algorithm (RGA) has never been used for solving the electromagnetic inverse problem. Practical experience and theoretical defense show that the RGA works better than the SGA, especially for problems with real design parameters. In this paper, the RGA is used to reconstruct the shape of a perfectly conducting cylinder in free space. Numerical results validate the algorithm. The advantages and disadvantages of RGA over gradient-based algorithms are discussed based on a comparison with the Newton-Kantorivitch method.