基于嵌入式菱形思维的参数实时评估微粒群算法

Real-time Parameter Estimation PSO Algorithm Based on Embedded Rhombus Thought

微粒群优化算法参数的选取是影响其性能和效率的关键。为了解决微粒群算法的参数优选问题,提出一种将可拓菱形思维嵌入到微粒群优化算法中,依赖被优化函数对参数进行自适应优选的高精度微粒群算法。基本思想是:(1)根据发散-收敛-再发散-再收敛这一菱形思维特点,建立微粒群算法参数优选的菱形思维模型,利用物元的可拓性对其进行发散性设计,再利用合适的评价方法对发散后的多种参数配置方案进行评价,取其中最优方案对应的参数作为算法当前代的参数值;(2)将菱形思维过程嵌入到微粒群算法的每一步,算法参数随着进化过程中的反馈信息不断被菱形思维优化,实现了参数选取过程的实时性和自适应性。该嵌入式优化算法既提高了算法的优化精度,又克服了迭代进化嵌套的高计算成本不足。最后通过对典型benchmark函数的优化仿真,表明该算法具有较高收敛速度和优化精度。

The choice of parameters proves to be crucial in enhancing the performance and efficiency of particle swarm optimization （PSO） algorithm. To provide good solution for reasonable choice of parameter values within fairly wide range for particle swarm optimization （PSO）, a novel parameter optimizing configuration strategy was proposed based on Embedded Rhombus Thought （ERT）, which depended on the optimization function to adaptively determine the appropriate set of parameters. The basic idea is： （1） a parameter optimization method of multi-stage rhombus thought model is established according to its divergent-concentrate-redivigent- reconcentrate nature, which carries on multiplicate sets of parameter values with divergent thought, and work out the optimum parameter values with proper convergent thought method; （2） the rhombus thought process is embedded in each PSO iteration, parameters are thus gradually optimized by the rhombus thought process as feedback information of the evolutionary process. The algorithm utilizes the advantages of the embedded optimization algorithm and overcomes its disadvantages, and experimental results on several benchmark function show that the algorithm can rapidly converge at very high quality solutions.