Parameters control in GAs for dynamic optimization

Abstract

The Control of Genetic Algorithms parameters allows to optimize the search process and improves the performance of the algorithm. Moreover it releases the user to dive into a game process of trial and failure to find the optimal parameters. Yet the control of parameters has received much attention in the case of static optimization problems, its investigation in the case of dynamic optimization problems (DOPs) is certainly a promising area of search. Indeed, in the case of DOPs the problem is not just to find the optima but to track the moving optima over time, so the parameters must be adapted to this dynamic environment. The proposed algorithm Parameters Control for Dynamic Optimization (PCDO) is based on Genetic Algorithm with Fitness Sharing (GAFS). To solve DOPs by controlling GAs parameters, PCDO uses several strategies. First, an unsupervised fuzzy clustering method is used to track multiple optimums and to perform GAFS. Second, a modified enthusiasm selection is used to adjust the selection pressure. Third, a clustering multi non uniform Mutation is utilized to locate an unexplored search space. Fourth, a novel technique with multiple crossover is applied to guide the algorithm in promising regions of the search space. Fifth, a self adaptive mutation rate is evolved through generation with a learned parameter, in order to control the diversity of the population. In the concern of maintaining the diversity of the population, a new genetic operator called Fertilization is proposed. PCDO is tested on six problems generated from Generalized Dynamic Benchmark Generator (GDBG). Experimental results demonstrate that PCDO outperforms other GAs on DOPs. Moreover, the ability of PCDO to maintain diversity is demonstrated by a new diversity measure.

Feedback-control operators for improved Pareto-set description: Application to a polymer extrusion process

“…It should be noticed here that the idea of feedback-control inspiration constitutes a further step beyond some studies that have considered the theme of parameter adaptation in evolutionary computation. Examples of those studies can be found for instance in Vasconcellos et al (2001), McGinley et al (2011), Lin and Chen (2013), and Jebari et al (2013). The introduction of feedback control concepts in order to state the parameter adaptation procedures allows the recovery, within the field of evolutionary computation, of some well-established results from control theory concerning closed-loop system dynamics and stability.…”

Neural Networks Applied in Chemistry. II. Neuro-Evolutionary Techniques in Process Modeling and Optimization

“…An initial population of potential solutions (in the applications we discuss, a "solution" is an ANN of value in solving a scientific problem) is iteratively improved through fitness-based selection and the generation of diversity. 15 Population-based EAs can create several elements of a Pareto optimal set in a single run when the chance of finding a unique optimal solution is low. 105 The GA, currently the most popular EA, was sketched out by Holland 16,17 and developed by numerous groups.…”