SEAL'06

Evolutionary Multiobjective Optimization and its Applications

Speaker: Gary G. Yen is a Professor in the School of Electrical and Computer Engineering, Oklahoma State University, USA. His research is supported by the DoD, DoE, EPA, NASA, NSF, and Process Industry. His research interest includes intelligent control, computational intelligence, conditional health monitoring, signal processing and their industrial/defense applications.

Dr. Yen was an associate editor of the IEEE Transactions on Neural Networks and IEEE Control Systems Magazine during 1994-1999. He is currently serving as an associate editor for the IEEE Transactions on Evolutionary Computation, IEEE Transactions on Control Systems Technology, IEEE Transactions on Systems, Man and Cybernetics and IFAC Journal on Automatica and Mechatronics. He served as the General Chair for the 2003 IEEE International Symposium on Intelligent Control held in Houston, TX and is the General Chair for the 2006 IEEE World Congress on Computational Intelligence to be held in Vancouver, Canada. On behalf of IEEE Robotic and Automation Society and later IEEE Control Systems Society, he was a Representative to the IEEE Neural Network Council Administrating Committee. Dr. Yen served as Vice President for the Technical Activities in IEEE Computational Intelligence Society in 2004 and 2005. He is the founding editor-in-chief of a newly founded IEEE/CIS Publication, IEEE Computational Intelligence Magazine.

Syllabus of the tutorial: Evolutionary computation is the study of biologically motivated computational paradigms which exert novel ideas and motivate inspiration from natural evolution and adaptation. The emerging use of evolutionary computation techniques has grown considerably over the past several years. During this period of time, the designs and applications of these techniques have been further enhanced resulting in a set of computational intelligence heuristics that are particularly adept for solving complex optimization problems. Moreover, they are characteristically more robust than traditional methods based on formal logics or mathematical programming in face of many real-world complications. Hence, evolutionary computation techniques have dealt with complex optimization problems better than traditional optimization techniques, although they can be applied to easy and simple problems with additional insights where conventional techniques fail to exhibit.

To its extreme, this population based approach with a little twist has been demonstrated to solve the multiobjective optimization problems efficiently and effectively. In many scientific and engineering disciplines, it is not uncommon to face a design challenge when there are several criteria or design objectives to be met simultaneously. If these objectives are conflicting, then the problem becomes one of finding the best possible designs that satisfy the competing objectives under different trade-off scenarios. With these multiple objectives and constraints taken into consideration, an optimum design problem in the context of Pareto optimality can then be formulated. Multiobjective Optimization Problem (MOP) is a very demanding research topic because most real-world problems have not only a multiobjective nature, but also many open issues to be answered qualitatively and quantitatively.

This tutorial will review state-of-the-art evolutionary computation techniques and its theoretical foundation and surveys the most recent developments in their use for solving complex multiobjective optimization problems and its applications. The outline of course syllabus will cover the following topics, but not limited to:

Evolutionary Computation for Real-World Problems

The number of possible solutions is so large that it precludes a complete search for the best answer. In other words, the number of possible distributions, routes, fraud rules, or transportation plans is so large, that examining all the possibilities would take many centuries of supercomputing time.

The problem exists in a time-changing environment. This means that yesterday’s decision, however optimal, may be far from optimal today.

The problem is heavily constrained. For most problems, the final solution should satisfy many restrictions imposed by internal regulations, capacities, laws, and/or preferences. Sometimes finding even one feasible solution (i.e., a solution that satisfies all problem-specific constraints) is quite difficult.

There are many (possibly conflicting) objectives. For example, the goal of many scheduling problems is to minimize both time and cost, but these two objectives work against each other (as a decrease in time usually results in an increase in cost, and vice versa). To allow business managers to effectively control these tradeoffs, such problems may require an entire set of solutions (rather than just a single solution).

In this tutorial we concentrate on constraint-handling techniques. We discuss many issues related to maintaining feasible and infeasible individuals in the population of potential solutions; in particular, we will discuss a variety of techniques based on:

Particle Swarm Optimization

Dr Xiaodong Li
School of Computer Science and IT, RMIT University
Email: xiaodong@cs.rmit.edu.au
http://goanna.cs.rmit.edu.au/~xiaodong/

Speaker: Dr Xiaodong Li is a Senior Lecturer in the School of Computer Science and IT, RMIT University, Melbourne, Australia. His research interest includes Artificial Intelligence, Evolutionary Computation, Artificial Neutral Networks, Swarm Intelligence, Multiobjective Optimization, Optimization in Dynamic Environments, and their applications to real-world problems. Dr. Li is an IEEE member, and a member of SIGEVO (The ACM Special Interest Group on Genetic and Evolutionary Computation). He serves as a technical committee member of Working Group on Swarm Intelligence, IEEE Computational Intelligence Society. He was the organizing chair for special session on Swarm Intelligence in CEC'03 and CEC'04, and again in CEC'06. He is the tutorial and special sessions chair for SEAL'06, and the publicity chair and a member of the steering committee for the IEEE Swarm Intelligence Symposium 2007 (SIS'07).

Syllabus of the tutorial: Particle Swarm Optimization (PSO) is a population-based stochastic optimization technique mimicking the social behaviours of animals and insects, such as bird flocking, animal herding, or fish schooling. PSO was first developed by James Kennedy and Russell Eberhart in 1995. In recent years it has gained increasing popularity in the Evolutionary Computation research community, largely due to the fact that PSO has been shown to be an effective optimization method for solving difficult optimization problems. PSO belongs to the family of Swarm Intelligence techniques, which typically involve studies of collective behaviour in decentralized systems. Such systems are made up by a population of simple agents interacting locally with one other and with their environment. Although there is typically no centralized control dictating the behaviour of the agents, local interactions among the agents often cause a global pattern to emerge. Swarm-like algorithms, such as (PSO) and Ant Colony Optimization (ACO), have already been applied successfully to solve real-world optimization problems.

PSO share some common characteristics with Evolutionary Algorithms. Like EAs, PSO starts with an initial population of randomly generated individuals (i.e., potential solutions). These individuals are then manipulated over many iterations by simulating the social behaviour of insects or animals, in an effort to find the optima in the problem space. Unlike EAs, PSO does not explicitly use evolutionary operators such as crossover and mutation. A potential solution simply 'flies' through the search space by modifying itself according to its past experience and its relationship with other individuals in the population and the environment.

This tutorial will present an introduction to Particle Swarm Optimization (PSO), and highlight some of the most important computational techniques employed and their recent development in this rapid growing area of research. Following topics will be covered:

Speaker: P. N. Suganthan received the B.A degree, Postgraduate Certificate and M.A degree in Electrical and Information Engineering from the University of Cambridge, UK. He obtained his PhD degree from the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. He was a pre-doctoral Research Assistant in the Department of Electrical Engineering, University of Sydney, Australia in 1995–96 and a lecturer in the Department of Computer Science and Electrical Engineering, University of Queensland, Australia in 1996–99. Since 1999 he has been with the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore where he was an Assistant Professor and now is an Associate Professor. He is an associate editor of the IEEE Transactions on Evolutionary Computation, Pattern Recognition Journal and International Journal of Computational Intelligence. His research interests include evolutionary computation, applications of evolutionary computation, pattern recognition, bioinformatics and neural networks. He is a senior member of the IEEE. He serves as a technical committee member of working group on swarm intelligence, IEEE Computational Intelligence Society. He is one of the organizers of CEC-05 special sessions on evolutionary real parameter optimization and CEC-06 special sessions on evolutionary constrained optimization. He has involved in the organization of over twenty conferences in various capacities. He has published over 100 technical articles in journals and conferences.

Syllabus of the tutorial: Although real parameter evolutionary algorithms have been around for some decades, the recently proposed Particle Swarm Optimizer (PSO) and Differential evolution (DE) have developed rapidly in recent years. PSO is based on swarm intelligence. It is somewhat different from other evolutionary algorithms as each particle in the swarm has a velocity and a pbest position to record and exploit its historical best positions of each particle. In addition, global best position, gbest is also remembered and exploited. Since Kennedy and Eberhart introduced it in 1995, it has attracted much attention and many research groups are actively working on it. Many interesting and improved variants of PSO have been developed based on Kennedy and Eberhart’s work and it has found numerous applications in many areas. DE algorithm, proposed by Storn and Price in 1995, is a simple but powerful population-based stochastic search technique for solving global optimization problems. Its efficiency has been successfully demonstrated in many application fields. DE’s mutation operators take the difference vector of two randomly chosen population vectors to perturb an existing vector. In this tutorial, we will present important recent developments of PSO & DE. In addition, the parameter adaptation in PSO & DE, multi-objective PSO & DE algorithms, constraint PSO & DE will also be discussed. The tutorial will also present our recent efforts to develop novel benchmark test functions to evaluate real parameter optimization algorithms. The problems of the existing test functions are discussed and novel single-objective, multi-objective and constrained test functions are suggested. The main topics of this tutorial are listed below:

Brief introduction to real-parameter evolutionary algorithms (e.g. DE, PSO, ...)
Recent variants of the PSO (e.g. CLPSO, DMS-PSO, parameter adaptation, ...)
Recent Variants of DE (e.g. self-adaptive DE, ...)
Development of test functions for global, constrained and multi-objective optimization
Experimental evaluation of DE & PSO variants on global, constrained and multi-objective optimization problems.