Effect of Learning Rate on the Performance of a PSS based PBIL

K.A. Folly (South Africa)

Keywords

Competitive learning, Genetic Algorithm, learning rate, premature convergence, PopulationBased Incremental Learning.

Abstract

A simple evolutionary algorithm called Population-Based Incremental Learning (PBIL) has been recently proposed as an alternative to Genetic Algorithm (GA) for the optimization of the parameters of power system damping controllers. PBIL is a technique that combines aspects of GAs with simple competitive learning. The advantage of PBIL over GA is that it is computationally simple and easy to use. The setting of the learning rate (LR) in PBIL can greatly affect the performance of the PBIL. The effect of the learning rate in PBIL is not well understood. This paper investigates the effect of learning rate (LR) on the performance of the PBIL. It is shown that on one hand, a small learning rate leads to more exploration of the search space and hence this introduces more diversity in the population. However, the convergence could be slow. On the other hand, a higher learning rate tends to lead to more exploitation of the information gained during the previous search and hence, a premature convergence could result.

Important Links:



Go Back