基于随机邻域变异和趋优反向学习的差分进化算法

Differential evolution algorithm based on random neighborhood mutation andoptimal opposition-based learning

传统差分进化(DE)算法在迭代过程中不能充分平衡全局勘探与局部开发,存在易陷入局部最优、求解精度低、收敛速度慢等缺点。为提升算法性能,提出一种基于随机邻域变异和趋优反向学习的差分进化(RNODE)算法并对其进行复杂度分析。首先,为种群中每个个体生成随机邻域,用全局最佳个体引导邻域最佳个体生成复合基向量,结合控制参数自适应更新机制构成随机邻域变异策略,使算法在引导种群向最优方向趋近的同时保持一定的勘探能力;其次,为了进一步帮助算法跳出局部最优,对种群中较差个体执行趋优反向学习操作,扩大搜索区域;最后,将RNODE与九种算法进行对比以验证RNODE的有效性和先进性。在23个Benchmark函数和两个实际工程优化问题上的实验结果表明,RNODE算法收敛精度更高、速度更快、稳定性更优。

The traditional differential evolution(DE)algorithm balanced global exploration and local exploitation inadequately,and had problems with easily falling into local optimal solutions,low solution accuracy and slow convergence speed.Therefore,this paper proposed a differential evolution algorithm based on random neighborhood mutation and optimal opposition-based learning(RNODE)and analyzed for its complexity.Firstly,the algorithm generated a random neighborhood for each individual in the current population,and used the global best individual to guide the neighborhood best individual to generate a composite basis vector,combined with an adaptive update mechanism of the control parameters to constitute a random neighborhood mutation strategy,which enabled the algorithm maintained its exploration ability and guided the population towards the optimal direction.Secondly,to further help the algorithm jump out of the local optimum,the algorithm performed the optimal opposition-based learning strategy on the poorer individuals to expand the search area.Finally,this paper compared RNODE with 9 algorithms to verify the effectiveness and advancement of RNODE.The experimental results on 23 benchmark functions and 2 real-world engineering optimization problems show that the RNODE algorithm has a higher convergence accuracy,faster speed and a greater stability.