Elitist Evolutionary Multi-Agent System in solving noisy multi-objective optimization problems

Abstract

Evolutionary Multi-Agent System approach for optimization (for multi-objective optimization in particular) is a promising computational model. Its computational as well as implemental simplicity cause that approaches based on EMAS model can be widely used for solving optimization tasks. It turns out that introducing some additional mechanisms into basic EMAS-such as presented in the course of this paper elitist extensions cause that results obtained with the use of proposed elEMAS (elitist Evolutionary Multi-Agent System) approach are as high-quality results as results obtained by such famous and commonly used algorithms as NSGA-II or SPEA2. Apart from the computational simplicity especially important and interesting aspects of EMAS-based algorithms it is characteristic for them a kind of soft selection which can be additionally easily adjusted depending on a particular situation-in particular it is possible to introduce auto-adapting selection into such systems. Such a kind of selection seems to be especially important and valuable in solving optimization tasks in uncertain or "noised" environments. In the course of this paper the model and experimental results obtained by elEMAS system in solving noisy multi-objective optimization problems are presented and the general conclusion is as follows: EMAS-based optimization system seems to be more effective alternative than classical (i.e. non agent-based) evolutionary algorithms for multi-objective optimization, in particular, in uncertain environment, it seems to be better alternative than NSGA-II algorithm.

Non-dominated Sorting Bee Colony optimization in the presence of noise

“…The 'soft' selection scheme of the multi-agent system proposed by Siwik and Natanek (2008) assists a quality solution (agent) to survive even with its poor fitness estimates in a noisy uncertain environment. Each solution (agent) here is assigned with a life energy which is reduced on relocating some resources to other dominating agents and vice versa.…”

“…The performance of the proposed noisy MOO algorithm realized with NSBC (hereafter referred to as Noisy Non-dominated Sorting Bee Colony, NNSBC) is compared with traditional NSBC, modified NSGA-II (Babbar et al 2003), NSGA-II with α-dominance operator (NSGA-II-A) (Boonma and Suzuki 2009), elitist Evolutionary Multi-Agent System (elEMAS) (Siwik and Natanek 2008), Noise-Tolerant Strength Pareto Evolutionary Algorithm (NT-SPEA) (Buche et al 2002), Differential Evolution for Multi-objective Optimization with Noise (DEMON) (Rakshit et al 2014), extended MOPSO (Rakshit et al 2014), extended NSGA-II (Rakshit et al 2014), and Differential Evolution for Multi-objective Optimization with Random Scale-Factor and Threshold Selection (DEMO-RSF-TS), which is an extended version of (Das et al 2005) to handle noise in MOO on the noisy version of a set of 23 benchmark functions (Zhang et al 2008). Experiments have also been undertaken to compare the performance of traditional DEMO, MOPSO and NSGA-II with their amended versions realized with the proposed noise handling strategies.…”

Differential evolution for noisy multiobjective optimization

“…( 1 2 ) The normalized measure of CD( i X G ), denoted by CD( ) i X G , is given in (13), where i X G and j X G lie in the same Pareto front:…”

“…The MOO algorithms used for the comparative study included DEMON [9], NSGA-II-A [10], CDR [11], simulated annealing for noisy MOO [12], elitist evolutionary multiagent system [13], MOEA-RF [14], modified NSGA-II [7], noise-tolerant strength Pareto evolutionary algorithm [15], and Pareto front-efficient global optimization [16]. These algorithms were selected for the comparative framework for their wide popularity in the realm of noisy MOO.…”

Agent-Based Multi-Objective Evolutionary Algorithms with Cultural and Immunological Mechanisms

“…• elitist and semi-elitist mechanisms (Siwik & Natanek, 2008b;Siwik & Natanek, 2008a;Siwik & Kisiel-Dorohinicki, 2006;Siwik & Kisiel-Dorohinicki, 2005 Siwik, 2006a;Drezėwski & Siwik, 2008a); During further research it has turned out that also mechanisms borrowed from immunological as well as from cultural algorithms can be introduced into EMAS and improve significantly its effectiveness taking solving MOOPs into account. One of possible realization of immune-cultural Evolutionary Multi-Agent System ic-EMAS (which in particular allows for overcoming mentioned in this section shortcomings of simple EMAS) is presented in the next section.…”