基于共享学习策略的微分进化算法

针对传统微分进化算法易发生早熟收敛问题，提出基于共享学习策略的微分进化算法（SLDE），引入共享个体和共享学习因子。共享个体覆盖整个种群，较优个体可引导算法朝希望方向进化，较差个体则能维持种群的多样性，向共享个体学习可避免丢失个体信息，实现整个种群间的信息交换，有助于算法跳出局部最优解，提高算法的局部开采和全局勘探能力。同时，算法充分利用个体的进化信息，根据个体适应值到最优适应值的距离自适应地调整共享学习因子，以弥补随机个体对进化带来的随机性和盲目性，增强算法的搜索能力。采用22个不同特性的Benchmark测试函数对算法进行性能测试，与7种改进DE算法进行性能对比，实验结果表明，SLDE具有较强的跳出局部最优解能力，能显著减少进化代数，大幅地提高算法的收敛精度、收敛速度和稳定性，SLDE的全局优化性能整体上远优于其他改进DE算法。

Differential Evolution Algorithm Based on Sharing Learning

In order to alleviate premature convergence in traditional differential evolution algorithms, a differential evolution algorithm based on sharing learning strategy (SLDE) was proposed and the concepts of sharing-individual (SI) and sharing learning factor were introduced in SLDE. The sharing-individual covers the whole population, while the superior individuals guide the promising searching direction, the inferior individuals maintain the population diversity in the evolution process. By learning from the sharing-individual, information exchange was achieved among the whole population to avoid missing information of individuals, which helps the algorithm jump over the trap of local optimal solution and improve the local and global exploration capability. Meanwhile, the evolutionary information of individuals was made full of use in SLDE, and the sharing learning factor was self-adaptively adjusted according to the distance of fitness value of individual and the optimal fitness value, in order to alleviate the randomness and blindness from the random individuals and enhance the searching ability. A total of 22 Benchmark test functions with different properties were used for performance test comparison with seven state-of-the-art DE variants. The experimental results showed that SLDE has strong ability to escape from local optima, significantly reduce the evolutionary generations and greatly improve the convergence precision, convergence speed and stability. The overall global optimization performance of SLDE is much better than other improved DE algorithms.