本論文研究存在於自由空間中二維金屬導體的頻域電磁逆散射問題。在正散射的分析部分，本研究以動差法(Method of Moments, MoM)為基礎，至於逆散射則被轉換為最佳化問題以進行求解，吾人使用非同步粒子群聚法(APSO)、非同步粒子群聚法結合平滑變化機制(APSO+Smoothvary)以及非同步粒子群聚法結合直交表和平滑變化機制(OA+APSO+Smoothvary)所求的解做比較。 本論文先探討非同步粒子群聚法、非同步粒子群聚法結合平滑變化機制與非同步粒子群聚法結合直交表和平滑變化機制在不同學習因子的情況下，以九種具不同特性之測試函數予以測試，維度方面皆設定為10維。每個測試函數最佳的學習因子和成功率不盡相同，經過實驗結果可以發現學習因子c1=2.8和c2=1.3是一組普遍通用的學習因子，在引進平滑變化機制之後，讓成功率大幅提昇之外，在收斂深度方面來的更深。 另外，直交表因為具有均勻分布的特性，將其結合並應用在逆散射問題時，可展現出其優越性，例如在初始最佳物種、收斂深度和圖形重建方面，皆有明顯的改善。為此，我們可看到針對二維金屬導體的逆散射問題，結合直交表及/或平滑變化機制之後皆可以有良好的改善。 This thesis studies the electromagnetic inverse scattering problem in frequency domain for a two-dimensional inhomogeneous dielectric cylinder located in free space. The analysis of forward scattering part is based on the Method of Moments (MoM) , while the inverse scattering part is tackled by transforming the problem into an optimization one, of which the asynchronous particle swarm optimization(APSO) is chosen. The reconstructed results by APSO are compared with those obtained by APSO plus certain kind of smooth variation for the control parameter and/or associated with the orthogonal array (OA).
At first, the convergence speed and results for nine benchmarked functions (with dimension 10) are tested as the control parameters are varied for the algorithm of APSO. It is found that not only the best values of the control parameters are different for each function, but also the best success rates are. Nevertheless, we do find one set of the learning factor, c1=2.8 and c2=1, that is suitable for all benchmarked functions tested in general. Then the introduction of certain kind of smooth variation for the control parameters is tested during the course of searching procedure. It is found that the mechanism of smooth variation for the control parameters can increase the success rate significantly in addition to the convergence depth. Finally, since the orthogonal array exhibits the characteristic of uniform appearance for each level of the experimental parameters, it do reveal its superiority as combined and applied for the inverse scattering problem. For examples, the initial best particle, convergence depth and reconstructed results can be significant improved. It is concluded that for the inverse scattering problem of a two-dimensional metallic conductor, the inclusion of orthogonal array and the mechanism of smooth variation of the control parameters is a helpful technique.