基于反馈策略的自适应粒子群优化算法

Adaptive particle swarm optimization algorithm based on feedback mechanism

为了克服常规粒子群优化(SPSO)算法在多峰函数寻优应用中容易出现早熟的缺点,提出了一种基于反馈策略的自适应粒子群优化(APSO)算法.考虑到进化过程中群体多样性损失过快,采用种群分布熵和平均粒距两个种群多样性参数,来均衡算法的勘探和开发能力.基于惯性权值随种群多样性变化而变化的动态分析,建立了惯性权值与平均粒距之间的线性函数关系,并将该函数关系融入到APSO算法中.测试结果表明,与常规粒子群优化算法相比,该算法在多峰雨数寻优时,成功率和精确度都有显著提高,且全局收敛速度快;在求解异或(XOR)分类问题时成功概率提高,收敛速度加快,APSO算法对神经网络的训练更加有效.

To overcome premature of multi-modal function search by standard particle swarm optimization （SPSO） algorithm, a new adaptive particle swarm optimization （APSO） based on feedback mechanism was proposed. Considering the large lost in population diversity during the evolution, two parameters of population-distribution-entropy and average-distance-amongst-points were introduced into the proposed algorithm to balance the trade-off between exploration and exploitation. A linear function relationship between inertia weight and average-distance-amongst-points was established through analyzing the dynamic relationship between inertia weight value and population diversity, and this functional relationship was embedded into APSO. The testing results indicate that APSO has better probability of finding global optimum, accuracy and speed of convergence than SPSO when APSO is applied to the solution of exclusive OR （XOR） classification problem, and that APSO is more efficient in training neural networks than in that of SPSO.