Advances in Swarm Intelligence

In this paper, we compare different aggregation strategies for cue-based aggregation with a mobile robot swarm. We used a sound source as the cue in the environment and performed real robot and simulation based experiments. We compared the performance of two proposed aggregation algorithms we called as the

vector averaging

and

naïve

with the state-of-the-art cue-based aggregation strategy BEECLUST. We showed that the proposed strategies outperform BEECLUST method. We also illustrated the feasibility of the method in the presence of noise. The results showed that the vector averaging algorithm is more robust to noise when compared to the naïve method.

The swarm behavior of pedestrians in a crowd, generally, causes a global pattern to emerge. A pedestrian crowd simulation system must have this emergence in order to prove its effectiveness. For this reason, the aim of our work is to demonstrate the effectiveness of our model PHuNAC (Personalities’ Human’s Nature of Autonomous Crowds) and also prove that the swarm behavior of pedestrians’ agents in our model allows the emergence of these global patterns. In order to validate our approach, we compared our system with real data. The conducted experiments show that the model is consistent with the various emergent behaviors and thus it provides realistic simulated pedestrian’s behavior.

According to benchmark learning theory in business management, a kind of competitive learning mechanism based on dynamic niche was set up. First, by right of imitation and learning, all the individuals within population were able to approach to the high yielding regions in the solution space, and seek out the optimal solutions quickly. Secondly, the premature convergence problem got completely overcame through new optimal solution policy. Finally, the algorithm proposed here is naturally adaptable for the dynamic optimization problems. The unique search model was analyzed and revealed in detail.

The 3-flip neighborhood local search (3FNLS) is an excellent heuristic algorithm for the set covering problem which has dominating performance on the most challenging crew scheduling instances from Italy railways. We introduce a method to further improve the effectiveness of 3FNLS by incorporating random flat move to its search process. Empirical studies show that this can obviously improve the solution qualities of 3FNLS on the benchmark instances. Moreover, it updates two best known solutions within reasonable time.

For hunting foods, migrating and breeding, some small animals tend towards flocking. Once encountering predators, these swarms usually present some special threat-evading behaviors and show incredible regularity in movement by coordinating with each other. In this paper, the actions of a small fish swarm evades a blacktip reef shark are taken as an example, we put forward the concept of predator threat-field and present a mathematical model to describe the threat-field intensity, demonstrate the decision-making process of preys escaping the predator’s threat by revising R-A model. The validity of the model is shown by simulations.

As a well-known NP-hard problem, the Three-Index Assignment Problem (AP3) has attracted lots of research efforts for developing heuristics. However, existing heuristics either obtain less competitive solutions or consume too much time. In this paper, a new heuristic named Approximate Muscle guided Beam Search (AMBS) is developed to achieve a good trade-off between solution quality and running time. By combining the approximate muscle with beam search, the solution space size can be significantly decreased, thus the time for searching the solution can be sharply reduced. Extensive experimental results on the benchmark indicate that the new algorithm is able to obtain solutions with competitive quality and it can be employed on instances with large-scale. Work of this paper not only proposes a new efficient heuristic, but also provides a promising method to improve the efficiency of beam search.

Fireworks algorithm (FWA) is a relatively new metaheuristic in swarm intelligence and EFWA is an enhanced version of FWA. This paper presents a new improved method, named IEFWA, which modifies EFWA in two aspects: a new Gaussian explosion operator that enables new sparks to learn from more exemplars in the population and thus improves solution diversity and avoids being trapped in local optima, and a new population selection strategy that enables high-quality solutions to have high probabilities of entering the next generation without incurring high computational cost. Numerical experiments show that the IEFWA algorithm outperforms EFWA on a set of benchmark function optimization problems.

Various metaheuristics have been proposed recently and each of them has its inherent evolutionary, physical-based, and/or swarm intelligent mechanisms. This paper does not focus on any subbranch, but on the metaheuristics research from a unified view. The population of decision vectors is looked on as an abstract matrix and three novel basic solution generation operations, E[p(i,j)], E[p(c·i)] and E[i, p(c·i+j)], are proposed in this paper. They are inspired by the elementary matrix transformations, all of which have none latent meanings. Experiments with real-coded genetic algorithm, particle swarm optimization and differential evolution illustrate its promising performance and potential.

Fruit fly optimization algorithm (FOA) was a novel swarm intelligent algorithm inspired by the food finding behavior of fruit flies. Due to the deficiency of trapping into the local optimum of FOA, a new fruit fly optimization integrated with chaos operation (named CFOA) was proposed in this paper, in which logistic chaos mapping was introduced into the movement of the fruit flies, the optimum was generated by both the best fruit fly and the best fruit fly in chaos. Experiments on single-mode and multi-mode functions show CFOA not only outperforms the basic FOA and other swarm intelligence optimization algorithms in both precision and efficiency, but also has the superb searching ability.

A new bio-inspired algorithm, Chicken Swarm Optimization (CSO), is proposed for optimization applications. Mimicking the hierarchal order in the chicken swarm and the behaviors of the chicken swarm, including roosters, hens and chicks, CSO can efficiently extract the chickens’ swarm intelligence to optimize problems. Experiments on twelve benchmark problems and a speed reducer design were conducted to compare the performance of CSO with that of other algorithms. The results show that CSO can achieve good optimization results in terms of both optimization accuracy and robustness. Future researches about CSO are finally suggested.

Extremal optimization is a dynamic, heuristic intelligent algorithm. It evolves a single solution and makes local modifications to the worst components. In this paper, a knowledge-base mutation operator is presented based on the distribution knowledge of candidate solutions. And then a population-based extremal optimization with knowledge-based mutation is proposed by introducing the idea of swarm evolution. Finally, the proposed method is applied to PID parameter tuning. The simulation results show that the proposed algorithm is characterized by high response speed, small overshoot and steady-state error, and obtains satisfactory control effect.

Magnetotactic bacteria is a kind of polyphyletic group of prokaryotes with the characteristics of magnetotaxis that make them orient and swim along geomagnetic field lines. Its distinct biology characteristics are useful to design new optimization technology. In this paper, a new bionic optimization algorithm named Magnetotactic Bacteria Moment Migration Algorithm(MBMMA) is proposed. In the proposed algorithm, the moments of a chain of magnetosomes are considered as solutions. The moments of relative good solutions can migrate each other to enhance the diversity of the MBMMA. It is compared with Genetic Algorithm, Differential Evolution and CLPSO on standard functions problems. The experiment results show that the MBMMA is effective in solving optimization problems. It shows good and competitive performance compared with the compared algorithms.

Magnetotactic bacteria is one kind of bacteria with magnetic particles called magnetosomes in its body. The magnetotactic bacteria move towards the ideal living conditions under the interaction between magnetic field produced by the magnetic particles chain and that of the earth. In the paper, a new magnetotactic bacteria algorithm based on power spectrum (PSMBA) for optimization is proposed. The candidate solutions are decided by power spectrum in the algorithm. Its performance is tested on 8 standard functions problems and compared with the other two popular optimization algorithms. Experimental results show that the PSMBA is effective in optimization problems and has good and competitive performance.

A proposal for particles’ initialization in PSO is presented and discussed, with focus on costly global unconstrained optimization problems. The standard PSO iteration is reformulated such that the trajectories of the particles are studied in an extended space, combining particles’ position and speed. To the aim of exploring effectively and efficiently the optimization search space since the early iterations, the particles are initialized using sets of orthogonal vectors in the extended space (orthogonal initialization, ORTHOinit). Theoretical derivation and application to a simulation-based optimization problem in ship design are presented, showing the potential benefits of the current approach.

The individual learning and team working is the quintessence of particle swarm optimization (PSO). Within the conceptual framework of computational thinking, the every particle is seen as a computing entity and the whole bird community is a generalized distributed, parallel, reconfigurable and heterogeneous computing system. Meanwhile, the small world network provides a favorable tool for the topology structure reconfiguration among birds. So a learning framework of distributed reconfigurable PSO with small world network (DRPSOSW) is proposed, which is supposed to give a systemative approach to improve algorithms. Finally, a series of benchmark functions are tested and contrasted with the former representative algorithms to validate the feasibility and creditability of DRPSOSW.

Particle Swarm Optimization (PSO) is popular in optimization problems for its quick convergence and simple realization. The topology of standard PSO is global-coupling and likely to stop at local optima rather than the global one. This paper analyses PSO topology with complex network theory and proposes two approaches to improve PSO performance. One improvement is PSO with regular network structure (RN-PSO) and another is PSO with random network structure (RD-PSO). Experiments and comparisons on various optimization problems show the effectiveness of both methods.

This paper presents a new variant of PSO, called fully learned multi-swarm particle swarm optimization (FLMPSO) for global optimization. In FLMPSO, the whole population is divided into a number of sub-swarms, in which the learning probability is employed to influence the exemplar of each individual and the center position of the best experience found so far by all the sub-swarms is also used to balance exploration and exploitation. Each particle updates its velocity based on its own historical experience or others relying on the learning probability, and the center position is also applied to adjust its flying. The experimental study on a set of six test functions demonstrates that FLMPSO outperform the others in terms of the convergence efficiency and the accuracy.

Circulant partial Hadamard matrices are useful in many scientific applications, yet the existence of such matrices is not known in general. Some Hadamard matrices of given orders are found via complete enumeration in the literature, but the searches are too computationally intensive when the orders are large. This paper introduces a search method for circulant partial Hadamard matrices by using natural heuristic algorithm. Slightly deviated from the swarm intelligence based algorithm, this method successfully generates a class of circulant partial Hadamard matrices efficiently. A table of circulant partial Hadamard matrices results are given at the end of this paper.

Travelling Salesman Problem (TSP) is a classical combinatorial optimization problem. This problem is NP-hard in nature and is well suited for evaluation of unconventional algorithmic approaches based on natural computation. Ant Colony Optimization (ACO) technique is one of the popular unconventional optimization technique to solve this problem. In this paper, we propose High Performance Ant Colony Optimizer (HPACO) which modifies conventional ACO. The result of implementation shows that our proposed technique has a better performance than the conventional ACO.

The solutions to Traveling Salesman Problem can be widely applied in many real-world problems. Ant colony optimization algorithms can provide an approximate solution to a Traveling Salesman Problem. However, most ant colony optimization algorithms suffer premature convergence and low convergence rate. With these observations in mind, a novel ant colony system is proposed, which employs the unique feature of critical tubes reserved in the

Physaurm

-inspired mathematical model. A series of experiments are conducted, which are consolidated by two real-world Traveling Salesman Problems. The experimental results show that the proposed new ant colony system outperforms classical ant colony system, genetic algorithm, and particle swarm optimization algorithm in efficiency and robustness.

To solve a travelling salesman problem by evolutionary algorithms, a challenging issue is how to identify promising edges that are in the global optimum. The paper aims to provide solutions to improve existing algorithms and to help researchers to develop new algorithms, by considering such a challenging issue. In this paper, three heuristic strategies are proposed for population based algorithms. The three strategies, which are based on statistical information of population, the knowledge of minimum spanning tree, and the distance between nodes, respectively, are used to guide the search of a population. The three strategies are applied to three existing algorithms and tested on a set of problems. The results show that the algorithms with heuristic search perform better than the originals.

We present a novel application of the Artificial Bee Colony algorithm to solve the non-unicost Set Covering Problem. The Artificial Bee Colony algorithm is a recent Swarm Metaheuristic technique based on the intelligent foraging behavior of honey bees. We present a 2-level metaheuristic approach where an Artificial Bee Colony Algorithm acts as a low-level metaheuristic and its paremeters are set by a higher level Genetic Algorithm.

Many different earning algorithms used for getting high performance in mathematics and statistical tasks. Recently, an Artificial Bee Colony (ABC) developed by Karaboga is a nature inspired algorithm, which has been shown excellent performance with some standard algorithms. The hybridization and improvement strategy made ABC more attractive to researchers. The two famous improved algorithms are: Guided Artificial Bee Colony (GABC) and Gbest Guided Artificial Bee Colony (GGABC), are used the foraging behaviour of the gbest and guided honey bees for solving optimization tasks. In this paper, GABC and GGABC methods are hybrid and so-called Hybrid Guided Artificial Bee Colony (HGABC) algorithm for strong discovery and utilization processes. The experiment results tested with sets of numerical benchmark functions show that the proposed HGABC algorithm outperforms ABC, PSO, GABC and GGABC algorithms in most of the experiments.

DNA microarrays are a powerful technique in genetic science due to the possibility to analyze the gene expression level of millions of genes at the same time. Using this technique, it is possible to diagnose diseases, identify tumours, select the best treatment to resist illness, detect mutations and prognosis purpose. However, the main problem that arises when DNA microarrays are analyzed with computational intelligent techniques is that the number of genes is too big and the samples are too few. For these reason, it is necessary to apply pre-processing techniques to reduce the dimensionality of DNA microarrays. In this paper, we propose a methodology to select the best set of genes that allow classifying the disease class of a gene expression with a good accuracy using Artificial Bee Colony (ABC) algorithm and distance classifiers. The results are compared against Principal Component Analysis (PCA) technique and others from the literature.

This work presents a crowding-distance(CD)-based multiobjective artificial bee colony algorithm for Proportional-Integral-Derivative (PID) parameter optimization. In the proposed algorithm, a new fitness assignment method is defined based on the nondominated rank and the CD. An archive set is introduced for saving the Pareto optimal solutions, and the CD is also used to wipe off the extra solutions in the archive. The experimental results compared with NSGAII over two test functions show its effectiveness, and the simulation results of PID parameter optimization verify that it is efficient for applications.

Concentration based feature construction (CFC) approach has been proposed for spam detection. In the CFC approach, Global concentration (GC) and local concentration (LC) are used independently to convert emails to 2-dimensional or 2n-dimensional feature vectors. In this paper, we propose a novel model which selects concentration construction methods adaptively according to the match between testing samples and different kinds of concentration features. By determining which concentration construction method is proper for the current sample, the email is transformed into a corresponding concentration feature vector, which will be further employed by classification techniques in order to obtain the corresponding class. The k-nearest neighbor method is introduced in experiments to evaluate the proposed concentration selection model on the classic and standard corpora, namely PU1, PU2, PU3 and PUA. Experimental results demonstrate that the model performs better than using GC or LC separately, which provides support to the effectiveness of the proposed model and endows it with application in the real world.

Immune control is a kind of intelligent control method which is based on biology immune system. It provides a new way for solving nonlinear, untertain and time variable system. In the paper, for the problem of Permanent Magnet Synchonous Motor (PMSM) speed control, an immune controller based on Varela immune network model is proposed. It uses immune feedback mechanism to construct the immune controller. It is compared with conventional PID controller for PMSM speed control. The Varela immune controller has a smaller starting current to avoid the influence of excessive current to PMSM. Simulation results show the effectiveness and practicability of this method.

In recent years, fuzzy job shop scheduling problems (FJSSP) with fuzzy triangular processing time and fuzzy due date have received an increasing interests because of its flexibility and similarity with practical problems. The objective of FJSSP is to maximize the minimal average customer’s degree of satisfaction. In this paper, a novel adaptive immune-genetic algorithm (CAGA) is proposed to solve FJSSP. CAGA manipulates a number of individuals to involve the progresses of clonal proliferation, adaptive genetic mutations and clone selection. The main characteristic of CAGA is the usage of clone proliferation to generate more clones for fitter individuals which undergo the adaptive genetic mutations, thus leading a fast convergence. Moreover, the encoding scheme of CAGA is also properly adapted for FJSSP. Simulation results based on several instances verify the effectiveness of CAGA in terms of search capacity and convergence performance.

As we know, genetic algorithm converges slowly. It is a natural contradiction when the situation appears with expensive objective function evaluating and satisfactory solutions being adequate. In this paper, a very fast convergent evolutionary algorithm (VFEA) is proposed with inner-outer hypercone crossover, problem dependent and search status involved mutation (PdSiMu). The offsprings produced by hypercone crossover are allowed to be outside the hypercone generated by rotating the parents around their bisectrix. PdSiMu utilizes the problem and evolving information quickly. VFEA is experimentally compared with five competitors based on ten classic 30 dimensional benchmarks. Experimental results indicate that VFEA can reach the accuracy of 10

− 4

 − 10

− 1

for all the benchmarks within 1500 function evaluations. VFEA arrives significantly better performance than all its competitors with higher solution accuracy and stronger robustness.

A variant of quantum evolutionary algorithm based on dynamic neighborhood topology(DNTQEA) is proposed in this paper. In DNTQEA, the neighborhood of a quantum particle are not fixed but dynamically changed, and the learning mechanism of a quantum particle includes two parts, the global best experience of all quantum particles in population, and the best experiences of its all neighbors, which collectively guide the evolving direction. The experimental results demonstrate the better performance of the DNTQEA in solving combinatorial optimization problems when compared with other quantum evolutionary algorithms.

A novel approach of function mining algorithm based on co-evolutionary gene expression programming (GEP-DE) which combines gene expression programming (GEP) and differential evolution (DE) was proposed in this paper. GEP-DE divides the function mining process of each generation into 2 phases: in the first phase, GEP focuses on determining the structure of function expression with fixed constant set, and in the second one, DE focuses on optimizing the constant parameters of the function which obtained in the first phase. The control experiments validate the superiority of GEP-DE, and GEP-DE performs excellently in the wheat aphid population forecast problem.

Accurate prediction of energy products future price is required for effective reduction of future price uncertainty as well as risk management. Neural Networks (NNs) are alternative to statistical and mathematical methods of predicting energy product prices. The daily prices of Propane (PPN), Kerosene Type Jet fuel (KTJF), Heating oil (HTO), New York Gasoline (NYGSL), and US Coast Gasoline (USCGSL) interrelated energy products are predicted. The energy products prices are found to be significantly correlated at 0.01 level (2-tailed). In this study, NNs learning algorithms are used to build a model for the accurate prediction of the five (5) energy product price. The aptitudes of the five (5) NNs learning algorithms in the prediction of PPN, KTJF, HTO, NYGSL, and USCGSL are examined and their performances are compared. The five (5) NNs learning algorithms are Gradient Decent with Adaptive learning rate backpropagation (GDANN), Bayesian Regularization (BRNN), Scale Conjugate Gradient backpropagation (SCGNN), Batch training with weight and bias learning rules (BNN), and Levenberg-Marquardt (LMNN). Results suggest that the LMNN and BRNN can be viewed as the best NNs learning algorithms in terms of R

2

and MSE whereas GDANN was found to be the fastest. Results of the research can be use as a guide to reduce the high level of uncertainty about energy products prices, thereby provide a platform for developmental planning that can result in the improvement economic standard.

This paper presents a multivariable grey neural network (MGM-NN) model for predicting the state of industrial equipments. It combines the merit of MGM model and RBF-NN model on time series forecast. This mode takes the dynamic correlations among multi variables and environment’s impact on state of equipment into consideration. The proposed approach is applied to the melt channel state forecast. The results are contrasted to MGM model executed on the same test set. The results show the accuracy and promising application of the proposed model.

The hydration of Portland cement is a complicated process and still not fully understood. Much effort has been accomplished over the past years to get the accurate model to simulate the hydration process. However, currently existing methods using positive derivation from the conditions for physical-chemical reaction are lack of information in real hydration data. In this paper, one model based on Flexible Neural Tree (FNT) with acceptable goodness of fit was applied to the prediction of the cement hydration process from the real microstructure image data of the cement hydration which has been obtained by Micro Computed Tomography (micro-CT) technology. Been prepared on the basis of previous research, this paper used probabilistic incremental program evolution (PIPE) algorithm to optimize the flexible neural tree structure, and particle swarm optimization (PSO) algorithm to optimize the parameters of the model. Experimental results show that this method is efficient.

For a fuzzy classifier automated design the hybrid self-configuring evolutionary algorithm is proposed. The self-configuring genetic programming algorithm is suggested for the choice of effective fuzzy rule bases. For the tuning of linguistic variables the self-configuring genetic algorithm is used. An additional feature of the proposed approach allows the use of genetic programming for the selection of the most informative combination of problem inputs. The usefulness of the proposed algorithm is demonstrated on benchmark tests and real world problems.

For the specific needs of the underwater unmanned vehicle (UUV) in the working environment, the autonomous suspending control method for UUV based on amendment of fuzzy control rules is advanced. According to the traditional fuzzy controller, this method based on particle swarm optimization (PSO) for online search strategies adjusts the amendment of fuzzy control rules in time. This online optimization of fuzzy control method has a faster convergence speed, and can effectively adaptive adjust the motion state of UUV, which carries out the autonomous suspending control within the scope of predetermined depth for the accuracy and robustness . It is illuminated by simulation experiments that this autonomous suspending control method for UUV based on amendment of fuzzy control rules is more effective against uncertain disturbance and serious nonlinear, time change process.

To acquire adaptive behaviors of multiple agents in the unknown environment, several neuro-fuzzy reinforcement learning systems (NFRLSs) have been proposed Kuremoto et al. Meanwhile, to manage the balance between exploration and exploitation in fuzzy reinforcement learning (FRL), an adaptive learning rate (ALR), which adjusting learning rate by considering “fuzzy visit value” of the current state, was proposed by Derhami et al. recently. In this paper, we intend to show how the ALR accelerates some NFRLSs which are reinforcement learning systems with a self-organizing fuzzy neural network (SOFNN) and different learning methods including actor-critic learning (ACL), and Sarsa learning (SL). Simulation results of goal-exploration problems showed the powerful effect of the ALR comparing with the conventional empirical fixed learning rates.

In this study, different methods entitled Centroidal Voronoi Tessellations and Levenberg-Marquardt applied on SP-QPSO separately to enhance its performance and discovering the optimum point and maximum/ minimum value among the feasible space. Although the results of standard SP-QPSO shows its ability to achieve the best results in each tested problem in local search as well as global search, these two mentioned techniques are applied to compare the performance of managing initialization part versus convergence of agents through the searching procedure respectively. Moreover, because SP-QPSO is tested on low dimensional problems in addition to high dimensional problems SP-QPSO combined with CVT as well as LM, separately, are also tested with the same problems. To confirm the performance of these three algorithms, twelve benchmark functions are engaged to carry out the experiments in 2, 10, 50, 100 and 200 dimensions. Results are explained and compared to indicate the importance of our study.

In this paper, we explore the potential of kernelized extreme learning machine (KELM) for efficient diagnosis of Parkinson’s disease (PD). In the proposed method, the key parameters in KELM are investigated in detail. With the obtained optimal parameters, KELM manages to train the optimal predictive models for PD diagnosis. In order to further improve the performance of KELM models, feature selection techniques are implemented prior to the construction of the classification models. The effectiveness of the proposed method has been rigorously evaluated against the PD data set in terms of classification accuracy, sensitivity, specificity and the area under the ROC (receiver operating characteristic) curve (AUC).

In this paper, an improved method for cancer classification based on particle swarm optimization (PSO) and priorinformation is proposed. Firstly, the proposed method uses PSO to implement gene selection. Then, the global search algorithm such as PSO is combined with the local search one such as backpropagation (BP) to model the classifier. Moreover, the prior information extracted from the data is encoded in PSO for better performance. The proposed approach is validated on two publicly available microarray data sets. The experimental results verify that the proposed method selects fewer discriminative genes with comparable performance to the traditional classification approaches.

This paper presents an improved Grover searching algorithm [1] which can auto-control the iterative processing when the number of target states is unknown. The final amplitude of the target states will be decreased if this number is unknown. So the question is how to perform Grover searching algorithm without the number of target states? As for this question, there are two conventional solutions. One solution tries to find the number of target states before performing the original algorithm. The other solution guesses a random k as the number of target states before performing the original algorithm. Both the two solutions need

$O(\sqrt{N})$

additional times Oracle calls than original algorithm and the answer of the first solution is non-deterministic while the second solution needs to check the correctness of the result. Assuming an operator which can judge the sign of the phases of superposition state, based on this technical, this paper shows a novel solution, which can perform Grover searching algorithm even if the number of target states is unknown. This solution only needs adding one gate, which can judge the sign of phase, and one more time Oracle call than the original algorithm.

In recent years, multi-objective optimization problems (MOPs) have attracted more and more attention, and various approaches have been developed to solve them. This paper proposes a new multi-objective evolutionary algorithm (MOEA), namely Pareto partial dominance on two selected objectives MOEA (PPDSO-MOEA), which calculates dominance between solutions using only two selected objectives when choosing parent population. In the proposed algorithm, two objectives are mainly selected with the first and the second largest distances to the corresponding dimension of the best point. PPDSO-MOEA switches the two-objective combination in every

I

g

generation to optimize all of the objective functions. The search performance of the proposed method is verified on many-objective 0/1 knapsack problems. State-of-the-art algorithms including plus .1em minus .1em PPD-MOEA, MOEA/D, UMOEA/D, and an algorithm selecting objectives with random method (RSO) are considered as rival algorithms. The experimental results show that PPDSO-MOEA outperforms all the four algorithms on most scenarios.

A simple, convenient and efficient multi-objective binary disperse optimized bacterial colony chemotaxis algorithm (MDOBCC) is proposed, in which the Disp(disperse update mechanism) is defined to handle 0-1 disperse optimization problems. The concept of chemotaxis center is proposed with the item of group and chemotaxis in order to improve the convergence rate of the algorithm. The definition of reference colony is used to retain the elite solution produced during the iteration; the definition of colony spatial radius and density is used to guide the bacteria for determinate variation, thus keeping the algorithm obtain even-distributed Pareto optimum solution set. Furthermore, the derivation analysis is given to prove the convergence of the algorithm and comes to the conclusion of global convergence. The simulate result confirmed the effectiveness of the algorithm.

Feature selection is an important data preprocessing technique in classification problems. This paper focuses on a new feature selection problem, in which sampling data of different features have different reliability degree. First, the problem is modeled as a multi-objective optimization. There two objectives should be optimized simultaneously: reliability and classifying accuracy of feature subset. Then, a multi-objective feature selection method based on particle swarm optimization, called JMOPSO, is proposed by incorporating several effective operators. Finally, experimental results suggest that the proposed JMOPSO is a highly competitive feature selection method for solving the feature selection problem with unreliable data.

Enhancing the convergence property is one of the main goals to achieve when designing a multi-objective particle swarm optimization (MOPSO) algorithm. To promote convergence, a turbulence mechanism derived from the bacteria quorum sensing behavior is introduced and a novel MOPSO (MOPSO-QSIT) is proposed. The inspired turbulence mechanism takes into effect only if the whole current population’ velocities are rather small (less than a predefined threshold), which enables to maintain the swarm diversity and avoids declining the swarm evolution. The MOPSO-QSIT algorithm has been tested on a set of benchmark functions and compared with other multi-objective optimization algorithms that are representative of the state-of-the-art. Simulation results illustrate that the proposed algorithm possesses the best convergence performance while keep good diversity performance, and is a competitively effective global optimization tool.

The problem of community structure discovery in complex networks has become one of the hot spots in recent years. This paper proposes a multiobjective genetic algorithm MOGCM to uncover community structure. This method overcomes the limitations of the community detection problems, choosing MinMaxCut and the community fitness as the objective functions. In the experiments, 2 well-known real-life networks are used to validate the performance and the results show that the method successfully detects the communities and it is competitive with state-of-the-art approaches.

This paper presents a new multi-objective jumping particle swarm optimization (MOJPSO) algorithm to solve the multi-objective multicast routing problem, which is a well-known NP-hard problem in communication networks. Each particle in the proposed MOJPSO algorithm performs four jumps, i.e. the inertial, cognitive, social and global jumps, in such a way, particles in the swarm follow a guiding particle to move to better positions in the search space. In order to rank the non-dominated solutions obtained to select the best guider of the particle, three different ranking methods, i.e. the random ranking, an entropy-based density ranking, and a fuzzy cardinal priority ranking are investigated in the paper. Experimental results show that MOJPSO is more flexible and effective for exploring the search space to find more non-dominated solutions in the Pareto Front. It has better performance compared with the conventional multi-objective evolutionary algorithm in the literature.

Physarum Polycephalum

is a primitive unicellular organism. Its foraging behavior demonstrates a unique feature to form a shortest path among food sources, which can be used to solve a maze. This paper proposes a

Physarum

-inspired multi-agent system to reveal the evolution of

Physarum

transportation networks. Two types of agents – one type for search and the other for convergence – are used in the proposed model, and three transition rules are identified to simulate the foraging behavior of

Physarum

. Based on the experiments conducted, the proposed multi-agent system can solve the two possible routes of maze, and exhibits the reconfiguration ability when cutting down one route. This indicates that the proposed system is a new way to reveal the intelligence of

Physarum

during the evolution process of its transportation networks.

This paper studies the preset time consensus problem of sing-integrator multi-agent systems for reaching the desired state in both undirected and directed communication networks. Two linear consensus protocols with time-varying gain are proposed. In fixed undirected networks, if the undirected topology is connected, the proposed protocol can achieve the consensus at the preset time even if only a portion of agents can obtain the desired state. In fixed directed networks, if the directed topology has a directed spanning tree, the proposed protocol can solve the consensus problem at a given preset time. Finally, numerical simulation results are presented to demonstrate the effectiveness of the theoretical results.

As the Internet become mainstream software system environment, software systems shift from closed, static and controllable to open, dynamic and difficult to control. The changes in the environment are unpredictable; it is major challenge for software system research to ensure that the software systems can deal with dynamic environment and change themselves appropriately. In this paper, according to the Multi-Agent environment, we divide environmental perception mechanism into three parts: by defining the environment, the composition problems in Multi-Agent Systems environment are solved; by designing method by which dynamic environmental data is generated and changes, we propose a dynamic environmental perception model based on the "publish / subscribe" model; by customized rules, the system can change itself in the environment according to the appropriate action to achieve the entire software adaptive process. Finally, we present examples to verify the feasibility and effectiveness of the theory.

Clustering of gene expression profiles is a mandatory task in cancer classification. Querying the expression of thousands of genes simultaneously imposes the use of powerful clustering techniques. Swarm based methods have shown their ability to perform data clustering. However, they may be faced to premature convergence problem and may be time consuming when large data sets need to be processed. Nowadays, the availability and widespread of parallel processing resources make possible the use of cooperative parallel methods. Within this context, we propose in this paper an archipelago based model that allows to reap advantage from the dynamics and the intrinsic parallelism of three swarm based methods namely PSO, ABC and ACO. Cooperation is achieved by sharing information through migration inside and between archipelagoes. The proposed cooperative parallel model for clustering gene expression profiles has been implemented on multicore computers and applied to several data sets. Experimental results show that it competes and even outperforms existing methods.

In view of the intrinsic drawbacks of traditional clustering methods, e.g. the sensitivity to initialization and the risk of falling into local optima, we introduce two new tools to enhance clustering performance via Swarm Intelligence (SI), i.e. Self-Aggregation (SA) and Eccentricity Analysis (EA), which are based on Firefly Algorithm (FA) in this paper. In order to confirm the effectiveness of the techniques, an improved k-means++ method is given as an instance. Large experiments illustrate that our algorithm performs better on both accuracy and robustness than the existing ones.

Using DNA computation to solve clustering problem is a new approach in this field. In the process of problem solving, we use DNA strands to assign vertices and edges, constructing the shortest Hamilton path and cutting branches whose length is longer than the threshold we gave getting the initial clustering result. For improving the quality, we do the iterative calculation, getting clusters for every produced cluster, we deal all of the process with DNA computation in test tubes, reducing the time complexity obviously by DNAs high parallelism. In this paper, we give the process and analysis of our algorithm, illustrating the feasibility of the method.

Cuckoo search (CS) is one of the new swarm intelligence optimization algorithms inspired by the obligate brood parasitic behavior of cuckoo, which used the idea of Lévy flights. But the convergence and stability of the algorithm is not ideal due to the heavy-tail property of Lévy flights. Therefore an improved cuckoo search (ICS) algorithm for clustering is proposed, in which the movement and randomization of the cuckoo is modified. The simulation results of ICS clustering method on UCI benchmark data sets compared with other different clustering algorithms show that the new algorithm is feasible and efficient in data clustering, and the stability and convergence speed both get improved obviously.

Clustering is a commonly used unsupervised machine learning method, which automatically organized data into different clusters according to their similarities. In this paper, we carried out a throughout research on evolutionary computation based clustering. This paper proposed a sample index based encoding method, which significantly reduces the search space of evolutionary computation so the clustering algorithm converged quickly. Evolutionary computation has a good global search ability while the traditional clustering method k-means has a better capability at local search. In order to combine the strengths of both, this paper researched on the effect of initializing k-means by evolutionary computation algorithms. Experiments were conducted on five commonly used evolutionary computation algorithms. Experimental results show that the sample index based encoding method and evolutionary computation initialized k-means both perform well and demonstrate great potential.

Artificial neural network (ANN) and support vector machine (SVM) based classifier design by a meta-heuristic called Co-Operation of Biology Related Algorithms (COBRA) is presented. For the ANN’s structure selection the modification of COBRA that solves unconstrained optimization problems with binary variables is used. The ANN’s weight coefficients are adjusted with the original version of COBRA. For the SVM-based classifier design the original version of COBRA and its modification for solving constrained optimization problems are used. Three text categorization problems from the DEFT’07 competition were solved with these techniques. Experiments showed that all variants of COBRA demonstrate high performance and reliability in spite of the complexity of the solved optimization problems. ANN-based and SVM-based classifiers developed in this way outperform many alternative methods on the mentioned benchmark classification problems. The workability of the proposed meta-heuristic optimization algorithms was confirmed.