Many, if not most, optimization problems have multiple objectives. Historically, multiple objectives have been combined ad hoc to form a scalar objective function, usually through a linear combination (weighted sum) of the multiple attributes, or by turning objectives into constraints. The genetic algorithm (GA), however, is readily modified to deal with multiple objectives by incorporating the concept of Pareto domination in its selection operator, and applying a niching pressure to spread its population out along the Pareto optimal tradeoff surface. We introduce the Niched Pareto GA as an algorithm for finding the Pareto optimal set. We demonstrate its ability to find and maintain a diverse "Pareto optimal population" on two artificial problems and an open problem in hydrosystems.

Niching methods extend genetic algorithms to domains that require the location and maintenance of multiple solutions. Such domains include classification and machine learning, multimodal function optimization, multiobjective function optimization, and simulation of complex and adaptive systems. This study presents a comprehensive treatment of niching methods and the related topic of population diversity. Its purpose is to analyze existing niching methods and to design improved niching methods. To achieve this purpose, it first develops a general framework for the modelling of niching methods, and then applies this framework to construct models of individual niching methods, specifically crowding and sharing methods. Using a constructed model of crowding, this study determines why crowding methods over the last two decades have not made effective niching methods. A series of tests and design modifications results in the development of a highly effective form of crowding, called determin...

In this paper, we study the problem features that may cause a multi-objective genetic algorithm (GA) difficulty in converging to the true Pareto-optimal front. Identification of such features helps us develop difficult test problems for multi-objective optimization. Multi-objective test problems are constructed from single-objective optimization problems, thereby allowing known difficult features of single-objective problems (such as multi-modality, isolation, or deception) to be directly transferred to the corresponding multi-objective problem. In addition, test problems having features specific to multiobjective optimization are also constructed. More importantly, these difficult test problems will enable researchers to test their algorithms for specific aspects of multi-objective optimization. Keywords Genetic algorithms, multi-objective optimization, niching, pareto-optimality, problem difficulties, test problems. 1 Introduction After a decade since the pioneering wor...

A technique is described which allows unimodal function optimization methods to be extended to efficiently locate all optima of multimodal problems. We describe an algorithm based on a traditional genetic algorithm (GA). This involves iterating the GA, but uses knowledge gained during one iteration to avoid re-searching, on subsequent iterations, regions of problem space where solutions have already been found. This is achieved by applying a fitness derating function to the raw fitness function, so that fitness values are depressed in the regions of the problem space where solutions have already been found. Consequently, the likelihood of discovering a new solution on each iteration is dramatically increased. The technique may be used with various styles of GA, or with other optimization methods, such as simulated annealing. The effectiveness of the algorithm is demonstrated on a number of multimodal test functions. The technique is at least as fast as fitness sharing methods. It provi...

Many, if not most, optimization problems have multiple objectives. Historically, multiple objectives (i.e., attributes or criteria) have been combined ad hoc to form a scalar objective function, usually through a linear combination (weighted sum) of the multiple attributes, or by turning objectives into constraints. The most recent development in the field of decision analysis has yielded a rigorous technique for combining attributes multiplicatively (thereby incorporating nonlinearity), and for handling uncertainty in the attribute values. But MultiAttribute Utility Analysis (MAUA) provides only a mapping from a vector-valued objective function to a scalar-valued function, and does not address the difficulty of searching large problem spaces. Genetic algorithms (GAs), on the other hand, are well suited to searching intractably large, poorly understood problem spaces, but have mostly been used to optimize a single objective. The direct combination of MAUA and GAs is a logical next step...

Researchers have long sought genetic algorithms (GAs) that can solve difficult search, optimization, and machine learning problems quickly. Despite years of work on simple GAs and their variants it is still unknown how difficult a problem simple GAs can solve, how quickly they can solve it, and with what reliability. More radical design departures than these have been taken, however, and the messy GA (mGA) approach has attempted to solve problems of bounded difficulty quickly and reliably by taking the notion of building-block linkage quite seriously. Early efforts were apparently successful in achieving polynomial convergence on some difficult problems, but the initialization bottleneck that required a large initial population was thought to be the primary obstacle to faster mGA performance. This paper replaces the partially enumerative initialization and selective primordial phase of the original messy GA with probabilistically complete initialization and a primordial phase that per...

This paper introduces and analyzes a parallel method of simulated annealing. Borrowing from genetic algorithms, an effective combination of simulated annealing and genetic algorithms, called parallel recombinative simulated annealing, is developed. This new algorithm strives to retain the desirable asymptotic convergence properties of simulated annealing, while adding the populations approach and recombinative power of genetic algorithms. The algorithm iterates a population of solutions rather than a single solution, employing a binary recombination operator as well as a unary neighborhood operator. Proofs of global convergence are given for two variations of the algorithm. Convergence behavior is examined, and empirical distributions are compared to Boltzmann distributions. Parallel recombinative simulated annealing is amenable to straightforward implementation on SIMD, MIMD, or shared-memory machines. The algorithm, implemented on the CM-5, is run repeatedly on two deceptive problems...

Genetic and evolutionary algorithms have been increasingly applied to solve complex, large scale search problems with mixed success. Competent genetic algorithms have been proposed to solve hard problems quickly, reliably and accurately. They have rendered problems that were difficult to solve by the earlier GAs to be solvable, requiring only a subquadratic number of function evaluations. To facilitate solving large-scale complex problems, and to further enhance the performance of competent GAs, various efficiency-enhancement techniques have been developed. This study investigates one such class of efficiency-enhancement technique called evaluation relaxation. Evaluation-relaxation schemes replace a high-cost, low-error fitness function with a low-cost, high-error fitness function. The error in fitness functions comes in two flavors: Bias and variance. The presence of bias and variance in fitness functions is considered in isolation and strategies for increasing efficiency in both cases are developed. Specifically, approaches for choosing between two fitness functions with either differing variance or differing bias values have been developed. This thesis also investigates fitness inheritance as an evaluation-

We approach the difficult task of analyzing the complex behavior of even the simplest learning classifier system (LCS) by isolating one crucial subfunction in the LCS learning algorithm: covering through niching. The LCS must maintain a population of diverse rules that together solve a problem (e.g., classify examples). To maintain a diverse population while applying the GA's selection operator, the LCS must incorporate some kind of niching mechanism. The natural way to accomplish niching in an LCS is to force competing rules to share resources (i.e., rewards). This implicit LCS fitness sharing is similar to the explicit fitness sharing used in many niched GAs. Indeed, the LCS implicit sharing algorithm can be mapped onto explicit fitness sharing with a one-to-one correspondence between algorithm components. This mapping is important because several studies of explicit fitness sharing, and of niching in GAs generally, have produced key insights and analytical tools for understanding th...

We assume that the modality (i.e., number of local optima) of a fitness landscape is related to the difficulty of finding the best point on that landscape by evolutionary computation (e.g., hillclimbers and genetic algorithms (GAs)). We first examine the limits of modality by constructing a unimodal function and a maximally multimodal function. At such extremes our intuition breaks down. A fitness landscape consisting entirely of a single hill leading to the global optimum proves to be hard for hillclimbers but apparently easy for GAs. A provably maximally multimodal function, in which half the points in the search space are local optima, can be easy for both hillclimbers and GAs. Exploring the more realistic intermediate range between the extremes of modality, we construct local optima with varying degrees of "attraction" to our evolutionary algorithms. Most work on optima and their basins of attraction has focused on hills and hillclimbers, while some research has explored attraction...