改进变异策略的自适应差分进化算法及其应用

针对标准差分进化(Differential Evolution,DE)算法存在控制参数难以选取以及变异策略缺乏稳健性的不足,提出一种基于改进变异策略的自适应DE算法.该方法用随机选取子种群的最优个体替代全种群最优个体进行变异操作;然后采用柯西分布和正态分布分别对收缩因子和交叉概率进行扰动,并在进化过程中利用胜出个体自适应地递推调整参数分布.对五个典型Benchmark函数的测试结果表明:该方法在全局搜索能力、收敛速度和精度以及可靠性上比标准DE算法都有了大幅度提高.将该方法用于分布式孔径雷达的方向图综合问题,通过较少的仿真次数即可稳定的获得更低的旁瓣电平.

Adaptive Differential Evolution Algorithm with Modified Mutation Strategy and Its Application

A modified mutation strategy based adaptive differential evolution (MMADE) algorithm is proposed aiming at the defects of standard DE which is lack of robustness and hard to choose the control parameters. MMADE is characterized by two factors: a new mutation strategy which utilizes the best individual of a randomly selected subgroup instead of the entire group and a parameter adaptation strategy for adjusting the parameters automatically. Mutation factors and crossover probabilities are generated randomly among each generation according to Cauchy and Gaussian distributions whose parameters are updated recursively by the winners of previous generation. Numerical simulations conducted by MMADE for five classic Benchmarks indicate greatly improved reliability, global search capability and convergence speed and accuracy compared with standard DE. Lower sidelobe level can be obtained through fewer simulations by applying MMADE to the pattern synthesis of distributed aperture radar.