Heterogeneous Vector-Evaluated Particle Swarm Optimisation in Static Environments
Particle swarm optimisation (PSO) is a population-based stochastic swarm intelligence (SI) optimization algorithm that converges very fast and thus lacks diversity. Heterogeneous vector evaluated particle swarm optimisation (HVEPSO) tries to introduce the
- PDF / 224,531 Bytes
- 12 Pages / 439.37 x 666.142 pts Page_size
- 82 Downloads / 200 Views
bstract. Particle swarm optimisation (PSO) is a population-based stochastic swarm intelligence (SI) optimization algorithm that converges very fast and thus lacks diversity. Heterogeneous vector evaluated particle swarm optimisation (HVEPSO) tries to introduce the ability to balance exploration and exploitation by increasing diversity of the particles’ behaviour. This study evaluates the performance of different HVEPSO configurations in static multi-objective environments. The particles of each sub-swarm of HVEPSO use different position and velocity update approaches selected from a behaviour pool. Strategies to determine when to change the particles’ behaviour are investigated for various knowledge transfer strategies (KTSs). Results indicate that the parent-centric crossover (PCX) KTS using the dynamic heterogeneous PSO (dHPSO) behaviour selection strategy with periodic window management performed the best. However, HVEPSO experienced problems converging to the optimal solutions and finding a diverse set of solutions for certain benchmarks, such as WFG1, which is a separable unimodal function with a convex Pareto-optimal front.
Keywords: Multi-objective optimisation
1
· Heterogeneous VEPSO
Introduction
Optimisation is the process of finding the best solution of a certain function, subject to given constraints [1]. Swarm intelligence (SI) algorithms are optimisation algorithms based on the study of individual behaviour in swarms or colonies. Particle swarm optimisation (PSO) is a population-based stochastic SI algorithm that is modelled on the social behaviour of bird flocks [4]. PSO was first introduced by Eberhart and Kennedy [3], where individuals are referred to as particles and the population is referred to as a swarm. Each particle’s behaviour is influenced by its position and velocity. When the particle’s behaviour (position and velocity) is updated, it causes the particle to move through the search space. Knowledge is kept on where the best position for the neighbourhood was (referred to as the neighbourhood best or nbest) and each particle’s position that leads to its best solution (referred to as the personal best or pbest) [4]. When all particles in a swarm use the same position and velocity update equations, the c Springer International Publishing Switzerland 2016 Y. Tan et al. (Eds.): ICSI 2016, Part I, LNCS 9712, pp. 293–304, 2016. DOI: 10.1007/978-3-319-41000-5 29
294
D. Doman et al.
PSO is referred to as a homogeneous PSO. However, if particles use different approaches to update their position and velocity, the PSO is referred to as a heterogeneous PSO (HPSO). Most real world problems are defined by multiple conflicting objectives and therefore require the simultaneous optimisation of a number of objectives. Parsopoulos and Vrahatis [18] developed the vector-evaluated particle swarm optimisation (VEPSO) algorithm based on the vector-evaluated genetic algorithm (VEGA) [19]. VEPSO uses sub-swarms, where each sub-swarm only solves one objective. Knowledge is then shared amongst the sub-swarms by using t
Data Loading...
