Innovations in Swarm Intelligence

In this chapter, advances in techniques and applications of swarm intelligence are presented. An overview of different swarm intelligence models is described. The dynamics of each swarm intelligence model and the associated characteristics in solving optimization as well as other problems are explained. The application and implementation of swarm intelligence in a variety of different domains are discussed. The contribution of each chapter included in this book is also highlighted.

Particle swarm optimization (PSO) is a metaheuristic inspired on the flight of a flock of birds seeking food, which has been widely used for a variety of optimization tasks [1,2]. However, its use in multimodal optimization (i.e., single-objective optimization problems having multiple optima) has been relatively scarce.

In this chapter, we will review the most representative PSO-based approaches that have been proposed to deal with multimodal optimization problems. Such approaches include the simple introduction of powerful mutation operators, schemes to maintain diversity that were originally introduced in the genetic algorithms literature (e.g., niching [3,4]), the exploitation of local topologies, the use of species, and clustering, among others.

Our review also includes hybrid methods in which PSO is combined with another approach to deal with multimodal optimization problems. Additionally, we also present a study in which the performance of different PSO-based approaches is assessed in several multimodal optimization problems. Finally, a case study consisting on the search of solutions for systems of nonlinear equations is also provided.

Swarm Intelligence is the part of Artificial Intelligence based on study of actions of individuals in various decentralized systems. The Bee Colony Optimization (BCO) metaheuristic has been introduced fairly recently as a new direction in the field of Swarm Intelligence. Artificial bees represent agents, which collaboratively solve complex combinatorial optimization problem. The chapter presents a classification and analysis of the results achieved using Bee Colony Optimization (BCO) to model complex engineering and management processes. The primary goal of this chapter is to acquaint readers with the basic principles of Bee Colony Optimization, as well as to indicate potential BCO applications in engineering and management.

The hypothesis that social influence leads to an increase in the division of labor, or specialization, in complex agent systems is introduced. Specialization, in turn, leads to increased productivity in such social systems. In this study, we examine the effect of social influence on the level of agent specialization in complex systems connected via social networks. Several methods attempt to explain the overall makeup of social influence and the emergence of specialization in general, with the most prominent being the genetic threshold model. This model posits that agents possess an inherent threshold for task stimulus, and when that threshold is exceeded, the agent will perform that task. The implication of social influence is that an agent’s choice of which task to specialize in when multiple ones are available is influenced by the choices of its neighbours. Using the threshold model and an established metric that quantifies the level of agent specialization, we find that social influence indeed leads to an increase in the division of labour.

We further investigate the sensitivity of the social influence rate on the overall level of system specialization. The social influence rate is important in the social context because it determines how swayed an agent is by the decisions of other agents in its group. On one hand, a high social influence rate will cause agents to mimic the collective behaviour of its neighbours. On the other hand, a rate that is too small will cause agents to choose primarily based on genetic factors. Experimental results, by way of comparing different rate selection strategies, reveal that increases in the social influence rate causes increases in the agent specialization within the system.

The computational complexity of ant colony optimization (ACO) is a new and rapidly growing research area. The finite-time dynamics of ACO algorithms is assessed with mathematical rigor using bounds on the (expected) time until an ACO algorithm finds a global optimum. We review previous results in this area and introduce the reader into common analysis methods. These techniques are then applied to obtain bounds for different ACO algorithms on classes of pseudo-Boolean problems. The resulting runtime bounds are further used to clarify important design issues from a theoretical perspective. We deal with the question whether the current best-so-far solution should be replaced by new solutions with the same quality. Afterwards, we discuss the hybridization of ACO with local search and present examples where introducing local search leads to a tremendous speed-up and to a dramatic loss in performance, respectively.

Many real-world problems involve optimization of multi-resolution parameters. In optimization problems, the higher the resolution, the larger the search space, and resolution affects the accuracy and performance of an optimization model. This article presents a genetic algorithm and particle swarm based cascade multi-resolution optimization model, and it is known as GA-PSO LECO. GA and PSO are combined in this research to integrate random as well as directional search to promote global exploration and local exploitation of solutions. The model is developed using the layered encoding representation structure, and is evaluated using two parameter optimization problems, i.e., the Tennessee Eastman chemical process optimization and the MMIC amplifier design interactive optimization.

Translational initiation sites (TISs) are an important type of gene signals which flag the starting location of the translation process. Due to the characteristics of the translational mechanism, an accurate recognition of TIS in a messenger RNA sequence leads to the determination of the primary structure of the corresponding protein. Many existing TIS prediction approaches investigate the data from one single perspective or apply some static central fusion mechanism on a fixed set of features. Due to the complicated nature of the genetic data, we believe that it is beneficial to consider multiple biological perspectives in the analysis process. In order to provide diversified problem solving techniques as well as modularization to the system, we have proposed a novel solution that uses a multi-agent system (MAS), which investigates the biological data from multiple biological perspectives, each of which is implemented by an independent problem solver agent with a unique expertise. A generalized layered framework is proposed to facilitate the system to take advantage of the synergy of having multiple agents working together and arriving at a single final prediction. In this chapter, we explore the application of particle swarm optimization and ant colony optimization in the proposed architecture. The integration of the swarm intelligent algorithms has lead to an outstanding performance of the system. Extensive experiments on three benchmark data sets have verified this claim, demonstrating the advantage of using the proposed system over most of the existing TIS prediction approaches.

In this chapter, three PSO based methods: Original PSO, Evolutionary PSO, and Adaptive PSO are compared for optimal operational planning problems of energy plants, which are formulated as Mixed-Intger Nonlinear Problems (MINLPs). The three methods are compared using typical energy plant operational planning problems. We have been developed an optimal operational planning and control system of energy plants using PSO (called FeTOP). FeTOP has been actually introduced and operated at three factories of one of the automobile company in Japan and realized 10% energy reduction compared with operators’ operation.

Advances in the development of nanotechnology gradually bring the field into its next generation involving systems of nanosystems. These bring about opportunities for computer science researchers to contribute their work as guidelines for the realisation and development of nanorobot systems in the near future. It is anticipated that an early version of future nanorobots may potentially contain only essential characteristics and exhibit only simple behaviours. It is similar to social insects in nature; collaborative behaviour among such simple individual exhibits a remarkable degree of intelligence. Hence, swarm intelligence techniques inspired by social insects could potentially be applied for nanorobot control mechanism in self-assembly. This study models an early version of future nanorobots and a control mechanism using swarm intelligence, especially PPSO (the modification of PSO for physical applications), for self-assembly and self-repair to examine the minimal characteristics and functionality for future nanorobots.

In the present study an ACO algorithm is adopted as a part of a document classification system that classifies documents written in Greek, in thematic categories. The main purpose of the ACO module is to create a word map that will assist in the representation of the documents in the pattern space. The word map creation algorithm proposed involves additional deterministic sub-routines and aims at clustering together into groups thematically-related words. The performance of the proposed system is compared with an alternative system implementation that is based on the established SOM neural network.

Identifying cancer biomarkers is an essential research problem that has attracted the attention of several research groups over the past decades. The main target is to find the most informative genes for predicting cancer cases, such genes are called cancer biomarkers. In this chapter, we contribute to the literature a new methodology that analysis the communities of genes to identify the most representative ones to be considered as biomarkers. The proposed methodology employs iterative t-test and singular value decomposition in order to produce the communities of genes which are analyzed further to identify the most prominent gene within each community; the latter genes are analyzed further as cancer biomarkers. The proposed methods have been applied on three microarray datasets. The reported results demonstrate the applicability and effectiveness of the proposed methodology.