In this talk, I will present a new probability learning based particle swarm optimization (PLPSO) algorithm for large-scale optimization. To make a good trade-off between learning efficiency and population diversity, PLPSO is built based on two novel probability learning schemas: elitist learning and social learning. The elitist learning, derived from the personal best learning of the standard PSO algorithm, enables each particle to learn from a randomly selected personal best position according to the learning probabilities. The social learning is a kind of population based learning, in which one particle learns from another particle of the current population, and the learning exemplar is also randomly selected according to the learning probabilities. Owing to the advantages of probability-based learning, both schemas enable particles to learn from "elites", but also ensure high degree of population diversities. Two open and widely used large-scale optimization benchmark function sets, CEC2010 and CEC2013, are employed for the performance testing. The empirical results show that, in comparison with five state-of-the-are schemes, PLPSO has both higher performance and less sensitivity to its key parameters.
Last modified: Tuesday, 19-Sep-2017 11:07:18 NZST
This page is maintained by the seminar list administrator.