《智能控制智能控制智能控制 (60).pdf》由会员分享,可在线阅读,更多相关《智能控制智能控制智能控制 (60).pdf(12页珍藏版)》请在得力文库 - 分享文档赚钱的网站上搜索。
1、Particle Swarm Optimization Bird Flocking A“swarm”is defined as an apparently disorganized collection(population)of moving individuals that tend to cluster together while each individual seems to be moving in a random direction.Bird flocking is one of the best examples of“swarm”in nature.From Bird t
2、o Particle Imagine a bird flock in an area where there is a single food source.A bird dont know where the food is,but it knows its distance to the food.The best strategy is to follow the bird that is closer to the food.Particles save and communicate the best solution they have found.From Bird to Par
3、ticle Population initialized by assigning random positions and velocities;potential solutions are then flown through hyperspace.Each particle keeps track of its“best”(highest fitness)position in hyperspace.“pBest”for an individual particle.“gBest”for the best in the population.“lBest”for the best in
4、 a defined neighborhood.At each time step,each particle randomly accelerates toward its pBest and gBest(or lBest).Process of PSO Step1.Initialize population in hyperspace.Step2.Evaluate fitness of individual particles.Step3.Modify velocities based on previous best and global(or neighborhood)best.Ste
5、p5.Go to step 2.Step4.Terminate on some condition.PSO Velocity Update Equations Global Version:dkxpcxpcvvtiktgktiktiktiktik,2,1,211dkvxxtiktiktik,2,1,11where k is the dimension,c1and c2are positive constants,and are random functions,and w is the inertia weight.For neighborhood version,change pgkto p
6、ik.Illustrating the Velocity Update Schema of Global PSOPSO:Related Issues Controlling velocities(determining the best value for Vmax)Usually set maximum velocity to dynamic range of variable Usually set c1and c2to 2 Inertia weight influences tradeoff between global and local exploration.Good approa
7、ch is to reduce inertia weight during run(i.e.,from 0.9 to 0.4 over 1000 generations)Swarm size and Neighborhood sizeAdvantages of PSO Adaptation operates on velocities Most other methods operate on positions Effect:PSO has a builtin momentum Particles tend to hurdle past optima an advantage,since t
8、he best positions are remembered anyway Simple in concept Easy to implement Computationally efficient Effective on a variety of problemsAn application of PSO Image Annotation RefinementPSO is used to optimize the feedforward neural network for combining the similarity measures between keywords.Y.Cao et al.,Using Neural Network to Combine Measures of WordSemantic Similarity for Image Annotation,ICIA 2011An application of PSO Image Annotation RefinementSummary Procedure of PSO An example Global PSO vs.Local PSO?