带有权重函数学习因子的粒子群算法

Particle swarm optimization algorithm with weight function's learning factor

粒子群算法(PSO)中惯性权重和学习因子的独自调整策略削弱了算法进化过程的统一性和粒子群的智能特性,很难适应复杂的非线性优化,为此提出一种利用惯性权重来控制学习因子的PSO算法。该算法将学习因子视作惯性权重的线性、非线性以及三角函数,在惯性权重随时间线性或非线性递减的过程中,学习因子发生相应的递减或递增变化,进而通过增强两者之间的相互作用来平衡算法的全局探索和局部开发能力,更好地引导粒子进行优化搜索。同时为了分析惯性权重和学习因子的融合性能,采用线性和非线性权重法进行比较,测试函数的优化结果表明了采用非线性递减权重的优越性。最后通过对多个基准测试函数的优化分析,并与带有异步线性变化和三角函数学习因子调整方法的PSO进行比较发现,该策略利用惯性权重调整学习因子,能达到平衡粒子个体学习能力和向群体学习能力的作用,提高了算法的优化精度。

Concerning the problem that the independent adjusting strategy of inertia weight and learning factor reduces evolution unity and intelligence feature of Particle Swarm Optimization（PSO） algorithm,and cannot adapt to the complex and nonlinear optimization problems,a new PSO algorithm with learning factor controlled by inertia weight function was proposed.Learning factor in the presented PSO was regarded as inertia weight＇s linear,nonlinear or trigonometric function,and increased or decreased progressively when inertia weight decreased by degrees linearly or nonlinearly.This strategy could effectively enhance the interaction of inertia weight and learning factor,then balance the global exploration and local exploitation and preferably lead particles to search globally optimal solution.Then the inertia weights were given to linear and nonlinear methods to analyze fusion performance of weight and learning factor,and the experimental results to test functions show that nonlinear weight is better.Finally,the experimental simulation results on benchmark test functions and the comparison with PSO with asynchronous linear and trigonometric function learning factor draw a conclusion that the strategy uses inertia weight to adjust learning factor,balances particle learning ability of individual and population,and improves optimization precision of algorithm.