Over the last three decades, a large number of evolutionary algorithms have been developed for solving multi-objective optimization problems. However, there lacks an upto-date and comprehensive software platform researchers to properly benchmark existing practitioners apply selected solve their real-world The demand such common tool becomes even more urgent, when source code many proposed has not made publicly available. To address these issues, we MATLAB in this paper, called PlatEMO, which...

Evolutionary algorithms often have to solve optimization problems in the presence of a wide range uncertainties. Generally, uncertainties evolutionary computation can be divided into following four categories. First, fitness function is noisy. Second, design variables and/or environmental parameters may change after optimization, and quality obtained optimal solution should robust against changes or deviations from point. Third, approximated, which means that suffers approximation errors....

In evolutionary multiobjective optimization, maintaining a good balance between convergence and diversity is particularly crucial to the performance of algorithms (EAs). addition, it becomes increasingly important incorporate user preferences because will be less likely achieve representative subset Pareto-optimal solutions using limited population size as number objectives increases. This paper proposes reference vector-guided EA for many-objective optimization. The vectors can used not...

In this paper, a novel competitive swarm optimizer (CSO) for large scale optimization is proposed. The algorithm fundamentally inspired by the particle but conceptually very different. proposed CSO, neither personal best position of each nor global (or neighborhood positions) involved in updating particles. Instead, pairwise competition mechanism introduced, where that loses will update its learning from winner. To understand search behavior theoretical proof convergence provided, together...

Evolutionary algorithms (EAs) have shown to be promising in solving many-objective optimization problems (MaOPs), where the performance of these heavily depends on whether solutions that can accelerate convergence toward Pareto front and maintaining a high degree diversity will selected from set nondominated solutions. In this paper, we propose knee point-driven EA solve MaOPs. Our basic idea is points are naturally most preferred among if no explicit user preferences given. A bias current...

Under mild conditions, it can be induced from the Karush-Kuhn-Tucker condition that Pareto set, in decision space, of a continuous multiobjective optimization problem is piecewise (m - 1)-D manifold, where m number objectives. Based on this regularity property, we propose model-based estimation distribution algorithm (RM-MEDA) for problems with variable linkages. At each generation, proposed models promising area space by probability whose centroid manifold. The local principal component...

It is not unusual that an approximate model needed for fitness evaluation in evolutionary computation. In this case, the convergence properties of algorithm are unclear due to approximation error model. paper, extensive empirical studies carried out investigate evolution strategy using function on two benchmark problems. found incorrect will occur if has false optima. To address problem, individual- and generation-based control introduced resulting effects presented. A framework managing...

During the past two decades, a variety of multiobjective evolutionary algorithms (MOEAs) have been proposed in literature. As pointed out some recent studies, however, performance an MOEA can strongly depend on Pareto front shape problem to be solved, whereas most existing MOEAs show poor versatility problems with different shapes fronts. To address this issue, we propose based enhanced inverted generational distance indicator, which adaptation method is suggested adjust set reference points...

Designing a controller for autonomous vehicles capable of providing adequate performance in all driving scenarios is challenging due to the highly complex environment and inability test system wide variety which it may encounter after deployment. However, deep learning methods have shown great promise not only excellent non-linear control problems, but also generalising previously learned rules new scenarios. For these reasons, use vehicle becoming increasingly popular. Although important...

The current literature of evolutionary many-objective optimization is merely focused on the scalability to number objectives, while little work has considered decision variables. Nevertheless, many real-world problems can involve both objectives and large-scale To tackle such (MaOPs), this paper proposes a specially tailored algorithm based variable clustering method. begin with, method divides variables into two types: 1) convergence-related 2) diversity-related Afterward, optimize types...

Most evolutionary optimization algorithms assume that the evaluation of objective and constraint functions is straightforward. In solving many real-world problems, however, such may not exist. Instead, computationally expensive numerical simulations or costly physical experiments must be performed for fitness evaluations. more extreme cases, only historical data are available performing no new can generated during optimization. Solving problems driven by collected in simulations,...

We propose a surrogate-assisted reference vector guided evolutionary algorithm (EA) for computationally expensive optimization problems with more than three objectives. The proposed is based on recently developed EA many-objective that relies set of adaptive vectors selection. (SAEA) uses Kriging to approximate each objective function reduce the computational cost. In managing models, focuses balance diversity and convergence by making use uncertainty information in approximated values given...

Evolutionary algorithms have been shown to be powerful for solving multiobjective optimization problems, in which nondominated sorting is a widely adopted technique selection. This technique, however, can computationally expensive, especially when the number of individuals population becomes large. mainly because most existing algorithms, solution needs compared with all other solutions before it assigned front. In this paper we propose novel, efficient approach sorting, termed sort (ENS)....

Federated learning obtains a central model on the server by aggregating models trained locally clients. As result, federated does not require clients to upload their data server, thereby preserving privacy of One challenge in is reduce client-server communication since end devices typically have very limited bandwidth. This article presents an enhanced technique proposing asynchronous strategy and temporally weighted aggregation local server. In strategy, different layers deep neural...

Fuzzy modeling of high-dimensional systems is a challenging topic. This paper proposes an effective approach to data-based fuzzy systems. An initial rule system generated based on the conclusion that optimal rules cover extrema. Redundant are removed similarity measure. Then, structure and parameters optimized using genetic algorithm gradient method. During optimization, have very low firing strength deleted. Finally, interpretability improved by fine training with regularization. The...

Constrained multiobjective optimization problems (CMOPs) are challenging because of the difficulty in handling both multiple objectives and constraints. While some evolutionary algorithms have demonstrated high performance on most CMOPs, they exhibit bad convergence or diversity CMOPs with small feasible regions. To remedy this issue, article proposes a coevolutionary framework for constrained optimization, which solves complex CMOP assisted by simple helper problem. The proposed evolves one...

Using surrogate models in evolutionary search provides an efficient means of handling today's complex applications plagued with increasing high-computational needs. Recent surrogate-assisted frameworks have relied on the use a variety different modeling approaches to approximate problem landscape. From these recent studies, one main research issue is choice scheme used, which has been found affect performance significantly. Given that theoretical knowledge available for making decision...

This paper investigates how to use prediction strategies improve the performance of multiobjective evolutionary optimization algorithms in dealing with dynamic environments. Prediction-based methods have been applied predict some isolated points both single objective and optimization. We extend this idea a whole population by considering properties continuous problems. In our approach, called strategy (PPS), Pareto set is divided into two parts: center point manifold. A sequence maintained...

In the real world, it is not uncommon to face an optimization problem with more than three objectives. Such problems, called many-objective problems (MaOPs), pose great challenges area of evolutionary computation. The failure conventional Pareto-based multi-objective algorithms in dealing MaOPs motivates various new approaches. However, contrast rapid development algorithm design, performance investigation and comparison have received little attention. Several test suites which were designed...

Machine learning is inherently a multiobjective task. Traditionally, however, either only one of the objectives adopted as cost function or multiple are aggregated to scalar function. This can be mainly attributed fact that most conventional algorithms deal with Over last decade, efforts on solving machine problems using Pareto-based optimization methodology have gained increasing impetus, particularly due great success evolutionary and other population-based stochastic search methods. It...

Function evaluations (FEs) of many real-world optimization problems are time or resource consuming, posing a serious challenge to the application evolutionary algorithms (EAs) solve these problems. To address this challenge, research on surrogate-assisted EAs has attracted increasing attention from both academia and industry over past decades. However, most existing (SAEAs) either still require thousands expensive FEs obtain acceptable solutions, only applied very low-dimensional In paper,...