粒子群优化算法的惯性权值递减策略研究

Study on the Strategy of Decreasing Inertia Weight in Particle Swarm Optimization Algorithm

为了有效地控制粒子群优化算法的全局搜索和局部搜索,基于递减惯性权值的基本思想,在现有的线性递减权值策略的基础上,提出了开口向下抛物线、开口向上抛物线和指数曲线3种非线性的权值递减策略,并采用Sphere、Rosenbrock、Griewank和Rastrigrin这4个标准测试函数测试这些策略对算法的影响.试验结果表明,对于多数连续优化问题,在初始权值和最终权值相同的情况下,凹函数递减策略优于线性策略,而线性策略优于凸函数策略,凹函数递减策略能够在不影响收敛精度的情况下较大幅度地提高粒子群算法的收敛速度.

To efficiently control the global and local search of particle swarm optimization （PSO）, motivated by the idea of decreasing inertia weight （DIW）, three nonlinear strategies for DIW, a parabola opening upwards, a parabola opening downwards and an exponential curve, are proposed based on the existing linear DIW. Sphere, Rosenbrock, Griewank and Rastrigrin functions are used to evaluate the strategies on the PSO performance. The experimental results show that for most continuous optimization problems, the strategy of concave function gains an advantage over the linear strategy, while the linear strategy outperforms strategy of convex function with the identical initial and final weights.