一种基于代理模型和差分进化的天线高维快速优化算法

An Efficient Optimization Method for High Dimensional Antenna Design Using Surrogate Model and Differential Evolution

天线结构日趋复杂,设计自由度不断提高,经典优化算法需要对大量的参数组合进行仿真试探后才能得到最优结果,使得天线的高维优化设计效率普遍较低.针对这一问题,将代理模型与进化算法相结合,提出基于Kriging模型的差分进化算法(Kriging based Differential Evolution Algorithm,KDEA).算法以Kriging模型部分代替电磁仿真,预测差分进化后个体的响应和不确定性;以进化前后种群的构成和筛选来调节搜索的广度和深度.结果显示,利用此方法优化一个9变量双层贴片天线的阻抗带宽及主瓣增益,相比同类优化算法,所需电磁仿真次数可以减少70%以上.

For novel antennas with complex structure and high degrees of freedom,the classical optimization methods require numerous simulation trials of different parameter combinations,which leads to a low efficiency in solving high dimensional antenna design and optimization problems.To address this issue,kriging based differential evolution algorithm（KDEA）,which combines evolution algorithm with surrogate model,is proposed.By partly substituting electromagnetic（EM）solver,kriging model predicts the responses and uncertainties of each individual after differential evolution.The exploration and exploitation of the search space can be adjusted by the constitution and prescreening of the population before and after evolution.This algorithm is applied to optimize the impedance bandwidth and the mainbeam gain of a stacked patch antenna with 9variables.Compared with the other optimization methods,the number needed for EM simulation reduced at least by 70%.