Evolving the Structure of Evolution Strategies using a Genetic Algorithm
نویسندگان
چکیده
Evolution strategies are one of the most successful classes of stochastic optimization algorithms for solving real world problems, which involves discontinuous, discrete or mixed-integer search space with nonlinear constraints. Since the creation of evolution strategies back in the 1960s, a variety of improvements and modifications have been suggested to enhance its performance. The covariance matrix adaptation evolution strategy (CMA-ES) is the state-of-the-art development in this category. Many variants have been proposed recently to accelerate its convergence speed. However, how to choose those variants optimally in practice is still an open question. In this thesis, based on the well-known No Free Lunch Theorem, we state that the optimal choice of variants should be related to the type of objective function landscape. For example, a certain variant favoring uni-modal landscape would perform worse on a multi-modal landscape. In order to obtain the optimal variant setting in practice, it is proposed to consider all the variants as a search space such that 1) many new combinations of variants beyond the literature can be tested and 2) an optimization algorithm can be used to search for the optimal ES-structure. An ES framework is presented in this thesis, which allows the usage of all the possible combinations of ES-variations. In addition, a genetic algorithm (GA) is exploited to evolve the ES-structure for a given black-box optimization problem using this framework. An empirical study is also conducted to validate the proposed approach, in which the GA is shown to converge fast and consistently to the best possible ES-structure within the framework by comparison with a brute force search. Performance of the evolved ES-structures is finally compared to the results of the black-box optimization benchmark (BBOB) from 2009 by means of the fixed cost error measure. Parts of this thesis are extracted from a paper by Van Rijn et al. [21].
منابع مشابه
Optimization of concrete structure mixture plan in marine environment using genetic algorithm
Today due to increasing development and importance of petroleum activities andmarine transport as well as due to the mining of seabed, building activities such as construction of docks, platforms and structures as those in coastal areas and oceans has increased significantly. Concrete strength as one of the most important necessary parameters for designing, depends on many factors such as mixtu...
متن کاملOPTIMAL DESIGN OF GRAVITY DAM USING DIFFERENTIAL EVOLUTION ALGORITHM
The shape optimization of gravity dam is posed as an optimization problem with goals of minimum value of concrete, stresses and maximum safety against overturning and sliding need to be achieved. Optimally designed structure generally saves large investments especially for a large structure. The size of hydraulic structures is usually huge and thus requires a huge investment. If the optimizatio...
متن کاملOptimization of Beam Orientation and Weight in Radiotherapy Treatment Planning using a Genetic Algorithm
Introduction: The selection of suitable beam angles and weights in external-beam radiotherapy is at present generally based upon the experience of the planner. Therefore, automated selection of beam angles and weights in forward-planned radiotherapy will be beneficial. Material and Methods: In this work, an efficient method is presented within the MATLAB environment to investigate how to improv...
متن کاملWell Placement Optimization Using Differential Evolution Algorithm
Determining the optimal location of wells with the aid of an automated search algorithm is a significant and difficult step in the reservoir development process. It is a computationally intensive task due to the large number of simulation runs required. Therefore,the key issue to such automatic optimization is development of algorithms that can find acceptable solutions with a minimum numbe...
متن کاملOptimal Design of Sandwich Panels Using Multi-Objective Genetic Algorithm and Finite Element Method
Low weight and high load capacity are remarkable advantages of sandwich panels with corrugated core, which make them more considerable by engineering structure designers. It’s important to consider the limitations such as yielding and buckling as design constraints for optimal design of these panels. In this paper, multi-objective optimization of sandwich panels with corrugated core is carried ...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2016