Swarm, Evolutionary, and Memetic Computing and Fuzzy and Neural Computing

A cooperative model of efficient evolutionary algorithms is proposed and studied when solving 22 real-world problems of the CEC 2011 benchmark suite. Four adaptive algorithms are chosen for this model, namely the Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES) and three variants of adaptive Differential Evolution (CoBiDE, jSO, and IDEbd). Five different combinations of cooperating algorithms are tested to obtain the best results. Although the two algorithms use constant population size, the proposed model employs an efficient linear population-size reduction mechanism. The best performing Cooperative Model of Evolutionary Algorithms (CMEAL) employs two EAs, and it outperforms the original algorithms in 10 out of 22 real-world problems.

In this article, we describe the design and implementation of a variant version of SOMA named SOMA Pareto to solve ten hard problems of the 100-Digit Challenge. The algorithm consists of the following operations: Organization, Migration, and Update. In which, we focus on improving the Organization operation with the adaptive parameters of PRT and Step. When applying the SOMA Pareto to solve ten hard problems to 10 digits of accuracy, we achieved a competitive result: 85.04 points.

In this paper we examined how the population size affects the performance of the differential evolution algorithm. First, we tested the original differential evolution algorithm, and then the improved self-adaptive differential evolution algorithm, on ten benchmark functions, that have been proposed for the CEC 2019 competition. We used six different population sizes. Afterwards, we tested the newly created algorithm with population reinitialization. The results show that the population size affects the algorithm’s efficiency, and that we need to tune it to obtain the best results. In the paper, we demonstrate that the newly created algorithm with reinitialization gives better, or at least comparable, results than the two algorithms without reinitialization.

In wireless communication arise various forms of optimization problems including channel assignment problem. There are many possible ways to assign channels to wireless links and our goal is to find the assignment that minimizes channel interference. For that purpose we have developed an ant colony optimization algorithm based on general guidelines of MAX-MIN Ant System and implemented it in C++ language. The algorithm was tested on problem instances and the results showed that the proposed algorithm is learning about instances that is solving and that way improves solution quality with the increase of iterations. Results confirmed that the proposed algorithm is an appropriate approach for solving channel assignment problem in cellular networks.

The self-organizing migrating algorithm (SOMA) is a popular population base metaheuristic. One of its key mechanisms is a perturbation of the individual movement with a binary-valued perturbation (PRT) vector. The goal of perturbation is to improve the diversity of the population and exploration of the search space. In this paper, we study a variant of the SOMA algorithm with non-binary PRT vector. We investigate the effect of introducing a third possible value, a negative (repulsive) element, into the PRT vector. The aim is to slow the population convergence and prolong the exploration phase. The inspiration is taken from previous successful implementations of repulsive mechanics in another swarm-based method: the Particle Swarm Optimization.

This paper is focused on the influence of boundary strategies for the popular swarm-intelligence based optimization algorithm: Self-organizing Migrating Algorithm (SOMA). A similar extensive study was already performed for the most famous representative of swarm-based algorithm, which is Particle Swarm Optimization (PSO), and showed the importance of related research for other swarm-based techniques, like SOMA. The current CEC’17 benchmark suite is used for the performance comparison of the case studies, and the results are compared and tested for statistical significance using the Friedman Rank test.

Pareto based selection techniques are extensively implemented in the multi-objective evolutionary algorithms (MOEAs), to tackle the many-objective optimization problems (MaOPs). In Pareto-dominance based MOEAs (PDMOEAs), nondominated sorting (NDS) plays a prominent role in preserving the elite solutions during mating and environmental selection. Although, NDS is an inevitable procedure in the evolution of PDMOEAs, computational complexity issues enhances the difficulty to adopt NDS approaches. Various methodologies were suggested in literature to overcome complexity issues, but these approaches deteriorate drastically for higher objectives. Recently, an approximate efficient NDS, (AENS) is proposed that utilize three objective comparisons to establish the dominance relation. In this paper, we propose an improved version of AENS, in which maximum two objective comparisons are required to determine the dominance relation. To evaluate the performance of our algorithm, experiments are done on seven different test problems and the experiment results have proved the effectiveness of proposed method in improving the convergence of different MOEAs.

The heat sink is one of the most widely used devices for thermal management of electronic devices and automotive systems. The present study approaches the design of the heat sink with the aim of enhancing their efficiency and keeping the material cost to a minimum. The above-mentioned purpose is achieved by posing the heat sink design problem as a bi-objective optimization problem where entropy generation rate and material cost are the two conflicting objective functions. The minimum entropy generation rate reduces irreversibilities inherent in the system, thus leading to improved performance, while the reduction in material cost ensures its economic feasibility. This bi-objective optimization problem is solved using Non-dominated Sorting Genetic Algorithm (NSGA-II) in the presence of geometric restrictions and functional requirements. Heat sinks with two different flow directions, namely flow-through air cooling system and impingement-flow air cooling system, are optimized to identify the best geometric and flow parameters. Subsequently a knowledge extraction exercise is carried out over non-dominated solutions obtained from the multi-objective optimization, to establish a relationship between the objective function and involved design parameters. The knowledge extracted has significant potential to simplify the calculations performed by thermal engineering experts in the selection of the heat sink for a specific application.

Predicting the trend of stock price movement accurately allows investors to maximize their profits from investments. However, due to the complexity of the stock data, classifiers often make errors, which cause the investors to lose money from failed investments. This study attempts to reduce such risks by focusing on easy-to-classify cases that have the highest chances of success. Therefore, we propose a method which selects only the predictions that have the highest confidence. In an experiment on 50 stocks, each learning model is trained on each stock data and evaluated based on the classification accuracy over a moving time window. The models which have the highest confidence are selected to predict the trend for that stock the next day. The experiment results shows the classification accuracy has improved significantly when the top 10% of predictions were used.

Aerofoil self-noise can affect the performance of the overall system. One of the main goals of aircraft design is to create an aerofoil with minimum weight, cost, and self-noise, satisfying all design requirements from the physical and the functional requirements. Aerofoil self-noise refers to the noise produced by the interaction between an aerofoil and its boundary layer. This paper describes how the prediction of the self-noise of an aerofoil at the early stage of the design phase can help select the best design of the aerofoil, which in turn reduces the lead time as the design process becomes more robust with respect to cost effectiveness. In the present work, the prediction of the self-noise of the aerofoil is addressed using Neural Networks (NN). Different architectures are used along with various proportions of training and testing to select the best architecture and best training-testing ratio. The results from NN is compared with linear, quadratic, and cubic polynomial regression. Thereafter, Principal Component Analysis (PCA) is integrated with NN for further improvement of prediction results. Our experimental results indicate that neural networks outperform regression. Moreover, PCA integrated with NN outperforms even the best neural network result.

A newly proposed variant of the efficient jSO algorithm employing competition of eight strategies (cSO) is proposed. The main idea is to select the most proper strategy and adapt the setting to each solved problem. One more mutation variant and one more type of crossover are added to jSO, and moreover, the popular mechanism of Eigen coordinate system is applied. All eight strategies compete to be used in the next generations based on the successes in previous generations. The proposed cSO method has more wins over jSO significantly in more real-world problems than fails. The original jSO strategy is never the most frequently used strategy, compared with other newly employed strategies.

The dramatic improvements in computational intelligence techniques over recent years have influenced many domains. Hence, it is reasonable to expect that virus writers will taking advantage of these techniques to defeat existing security solution. In this article, we outline a possible dynamic swarm smart malware, its structure, and functionality as a background for the forthcoming anti-malware solution. We propose how to record and visualize the behavior of the virus when it propagates through the file system. Neural swarm virus prototype, designed here, simulates the swarm system behavior and integrates the neural network to operate more efficiently. The virus’s behavioral information is stored and displayed as a complex network to reflect the communication and behavior of the swarm. In this complex network, every vertex is then individual virus instances. Additionally, the virus instances can use certain properties associated with the network structure to discovering target and executing a payload on the right object.

Knowledge discovery in databases is a comprehensive procedure which enables researchers to explore knowledge and information from raw sample data usefully. Some problems may arise during this procedure, for example the Curse of Dimensionality, where the reduction of database is desired to avoid feature redundancy or irrelevancy. In this paper, we propose a wrapper-based feature selection algorithm, consisting of an artificial neural network and self-adaptive differential evolution optimization algorithm. We test performance of the feature selection algorithm on a case study of bank marketing and show that this feature selection algorithm reduces the size of the database and simultaneously improves prediction performance on the observed problem.

In this paper, we address 10 basic test functions of the 100-Digit Challenge of the SEMCCO 2019 & FANCCO 2019 Competition by using team-to-team adaptive seft-organizing migrating algorithm - SOMA T3A with many improvements in the Organization, Migration, and Update process, as well as the linear adaptive PRT and the cosine-based adaptive Step. The results obtained from the algorithm on the 100-Digit Challenge are very promising with 93 points in total.

The success of every stochastic population-based nature-inspired algorithms is characterized through the dichotomy of exploration and exploitation. In general, exploration refers to the evaluation of points in previously untested areas of a search space, while exploitation refers to evaluation of points in close vicinity to previously visited points. How to balance both components properly during the evolutionary process is still considered as a topical problem in the evolutionary computation community. In this paper, we propose a recurrence plot visualization method for evaluating this process. Our analysis shows that recurrence plots are highly appropriate for revealing how particular algorithms balance exploration and exploitation.

Industrial manipulators are robots used to replace humans in dangerous or repetitive tasks. Also, these devices are often used for applications where high precision and accuracy is required. The increase of backlash caused by wear, that is, the increase of the amount by which teeth space exceeds the thickness of gear teeth, might be a significant problem, that could lead to impaired performances or even abrupt failures. However, maintenance is difficult to schedule because backlash cannot be directly measured and its effects only appear in closed loops. This paper proposes a novel technique, based on an Evolutionary Algorithm, to estimate the increase of backlash in a robot joint transmission. The peculiarity of this method is that it only requires measurements from the motor encoder. Experimental evaluation on a real-world test case demonstrates the effectiveness of the approach.

Cloud computing is rather important distributing computing paradigm and in general refers to the common pool of configurable resources that is accessed on-demand. Resources are dynamically scalable and metered with the basic aim to provide reliable and quality services to the end-users. Load scheduling has a great impact on the overall performance of the cloud system, and at the same time it is one of the most challenging problems in this domain. In this paper, we propose implementation of the hybridized elephant herding optimization applied to load scheduling problem in cloud computing. The algorithm is using CloudSim framework, and comparison with different metaheuristics, adapted and tested under same experimental conditions, for this type of problem was performed. Moreover, we compared proposed hybridized elephant herding optimization with its original version in order to evaluate its improvements in performance over the original version. Obtained empirical results prove the robustness and quality of approach that we propose in this paper.