Soft Computing in Industrial Applications

This paper provides a discussion on the main changes required in order to extend particle swarm optimization to the solution of multi-objective optimization problems. A short discussion of some potential paths for future research in this area is also included.

The judicious use of end-use electric loads in the framework of demand-side management programs as valuable resources to increase the global efficiency of power systems as far as economical, technical and quality of service aspects are concerned, is a relevant issue in face of the changes underway in the power systems industry. This paper presents the results of a multi-objective optimization model, in the perspective of an electricity retailer, which is aimed at designing load control actions to be applied to groups of electric loads. An evolutionary algorithm is used to compute solutions to this problem.

Many problems can be formulated as optimisation problems. Among the many classes of algorithms for solving such problems, one interesting, biologically inspired group is that of evolutionary optimisation techniques. In this tutorial paper we provide an overview of such techniques, in particular of Genetic Algorithms and Genetic Programming and its related subtasks of selection, cross-over, mutation, and coding. We then also explore Ant Colony Optimisation and Particle Swarm Optimisation techniques.

Grammar Induction (or Grammar Inference or Language Learning) is the process of learning of a grammar from training data of the positive and negative strings of the language. Genetic algorithms are amongst the techniques which provide successful result for the grammar induction. The paper is an extended approach to the earlier work by the authors regarding using stochastic mutation scheme based on Adaptive Genetic Algorithm for the induction of the grammar. Optimization by Genetic Algorithm often comes with premature convergence. The paper suggests two approaches, Elite Mating Pool and generating the population with the Dynamic Application of Reproduction Operator, for handling local convergence by considering a set of eleven different languages and their comparison. The algorithm produces successive generations of individuals, computing their ‘fitness value’at each step and selecting the best of them when the termination condition is reached. The paper deals with the issues in implementation of the algorithm,chromosome representation and evaluation, selection and replacement strategy, and the genetic operators for crossover and mutation. The model has been implemented, and the results obtained for the set of eleven languages are shown in the paper.

Quantum Evolutionary Algorithms (QEA) are novel algorithms proposed for class of combinatorial optimization problems. The probabilistic representation of possible solutions in QEA helps the q-individuals to represent all the search space simultaneously. In QEA, Q-Gate plays the role of update operator and moves qindividuals toward better parts of search space to represent better possible solutions with higher probability. This paper proposes an alternative magnetic update operator for QEA. In the proposed update operator the q-individuals are some magnetic particles attracting each other. The force two particles apply to each other depends on their fitness and their distance. The population has a cellular structure and each q-individual has four neighbors. Each q-individual is attracted by its four binary solution neighbors. The proposed algorithm is tested on Knapsack Problems, Trap problem and fourteen numerical function optimization problems. Experimental results show better performance for the proposed update operator than Q-Gate.

Population-based incremental learning (PBIL) is one of the well-established evolutionary algorithms (EAs). This method, although having outstanding search performance, has been somewhat overlooked compared to other popular EAs. Since the first version of PBIL, which is based on binary search space, several real code versions of PBIL have been introduced; nevertheless, they have been less popular than their binary code counterpart. In this paper, a population-based incremental learning algorithm dealing with real design variables is proposed. The method achieves optimization search with the use of a probability matrix, which is an extension of the probability vector used in binary PBIL. Three variants of the new real code PBIL are proposed while a comparative performance is conducted. The benchmark results show that the present PBIL algorithm outperforms both its binary versions and the previously developed continuous PBIL. The new methods are also compared with well-established and newly developed EAs and it is shown that the proposed real-code PBIL can rank among the high performance EAs.

Particle Swarm Optimization (PSO) is a popular optimization technique based on swarm intelligence concepts. Estimation of Distribution Algorithms (EDAs) are a relatively new class of evolutionary algorithms which build a probabilistic model of the population dynamics and use this model to sample new individuals. Recently, the hybridization of PSO and EDAs is emerged as a new research trend. In this paper, we introduce a new hybrid approach that uses a mixture of Gaussian distributions. The obtained algorithm, called PSEDA, can be seen as an implementation of the PSO behaviour in the EDAs framework. Experiments on well known benchmark functions have been held and the performances of PSEDA are compared with those of classical PSO.

The Generalised Nash Equilibrium Problem (GNEP) is a Nash game with the distinct feature that the feasible strategy set of a player depends on the strategies chosen by all her opponents in the game. This characteristic distinguishes the GNEP from a conventional Nash Game. These shared constraints on each player’s decision space, being dependent on decisions of others in the game, increases its computational difficulty. A special solution of the GNEP is the Nash Normalized Equilibrium which can be obtained by transforming the GNEP into a bi-level program with an optimal value of zero in the upper level. In this paper, we propose a Differential Evolution based Bi-Level Programming algorithm embodying Stochastic Ranking to handle constraints (DEBLP-SR) to solve the resulting bi-level programming formulation. Numerical examples of GNEPs drawn from the literature are used to illustrate the performance of the proposed algorithm.

In this paper, a new method for estimation of CO conversion in a range of temperatures, pressures and H2/CO molar ratios in the Fischer-Tropsch (FT) synthesis based on Locally Liner Model Tree (LoLiMoT) has been introduced. LoLiMoT is an incremental tree-construction algorithm that partitions the input space by axis-orthogonal splits. In each iteration two new local models as the result of splitting the worst local model has been inserted into the previous structure and result decreasing the total error. The system has been evaluated through two methods and results show estimated CO conversion values by LoLiMoT are in good agreement with experimental data.

This work proposes a multi-start global optimization algorithm that uses dimensional reduction techniques based upon approximations of space-filling curves and simulated annealing, aiming to find global minima of real-valued (possibly multimodal) functions that are not necessarily well behaved, that is, are not required to be differentiable or continuous. Given a real-valued function with a multidimensional and compact domain, the method builds an equivalent, onedimensional problem by composing it with a space-filling curve (SFC), searches for a small group of candidates and returns to the original higher-dimensional domain, this time with a small set of “promising” starting points. Finally, these points serve as seeds to the algorithm known as Fuzzy Adaptive Simulated Annealing, aiming to find the global optima of the original cost functions. New SFCs are built with basis on the well-known Sierpiński SFC, a subtle modification of a theorem by Hugo Steinhaus and several results of ergodic theory.

For a variety of practical uses, modelling techniques are being building up with the endeavor of reducing the expenditure and time related with the improvement of new Coriolis mass flow sensors [CMFS]. In this paper the phase shift which is linearly proportional to mass flow rate is modeled using an ANFIS. This technique is competent of understanding an immense diversity of non-linear correlations of substantial intricacy. The experimental data obtained from experimentation on indigenously developed Copper CMFS test rig is used for training the Anfis model then this model is accessible to the network in the structure of input-output pairs, thus the best possible correlation is found between the phase shift and influential important parameters. The training data is having phase shift at changeable input factors like sensor location, drive frequency and mass flow rate. Further, the multilayer feed forward neural network (MFNN) model is developed and compared with the ANFIS model results. These results reveal that ANFIS models could be effectively used in the expansion of Copper Coriolis mass flow sensors.

This paper proposes the tuning of a class of fuzzy control systems to ensure a reduced parametric sensitivity on the basis of a new Gravitational Search Algorithm (GSA). The GSA is employed to solve the optimization problems characterized by the minimization of objective functions defined as integral quadratic performance indices. The performance indices depend on the control error and on the squared output sensitivity functions of the sensitivity models with respect to the parametric variations of the controlled process. The controlled processes in the fuzzy control systems are benchmarks modeled by second-order linearized systems with an integral component and Takagi-Sugeno proportional-integral fuzzy controllers are designed and tuned for these processes.

In consumer-to-consumer (C2C) e-commerce environments, the magnitude of products and the diversity of vendors have caused confusion and difficulty for consumers to choose the right product from a trustworthy vendor. Feedback system is a widely used solution to help consumers evaluate vendors’ reputations. Some C2C environments have started to provide detailed feedback besides the overall rating system to help consumers distinguish individual vendors from multiple aspects. However, the increase in detailed feedback may add to consumer confusion and increase the time needed to consider all aspects for a reputation evaluation decision. This paper analyzes a typical feedback and reputation system for the e-commerce environment and proposes a novel, perception-based reputation model for individual vendors.

Biofeedback has been identified to improve postural control and stability. A biofeedback system communicates with the humans’ Central Nervous System through many available modalities, such as vibrotactile. The vibrotactile nature of feedback is presented in a simple and realistic manner, making the presentation of signals safe and easy to decipher. This work presents a wobble board training routine for rehabilitation combined with real-time biofeedback. The biofeedback was stimulated using a fuzzy inference system. The fuzzy system had two inputs and one output.Measurements to test this rehabilitation approach was taken in Eyes Open and Eyes Close states, with and without biofeedback while subjects stood on the wobble board. An independent

T

-test was conducted on the readings obtained to test for statistical significance. The goal of this work was to determine the feasibility of implementing a negative close-loop biofeedback system to assist in proprioceptor training utilizing wobble boards.

Today, wireless sensor networks encompass a large volume of applications. Wireless sensor networks consisted of many nodes by low energy batteries. Therefore, they must consume power as low as possible. TDMA Protocol in these networks is designed for this goal. In this paper a multiobjective immune algorithm is proposed for finding optimal solutions to TDMA scheduling problem. The simulation results show a better performance in comparison to two algorithms using instances with different sizes.

The generalized minimum spanning tree problem is a natural extension of the classical minimum spanning tree problem, looking for a tree with minimum cost, spanning exactly one node from each of a given number of predefined, mutually exclusive and exhaustive node sets. In this paper we present a memetic algorithms for solving the generalized minimum spanning tree problem that combines the population concept of genetic algorithms with a fast local improvement method. The proposed algorithm is competitive with other heuristics published to date in both solution quality and computation time. The computational results for several benchmarks problems are reported and the results point out that the memetic algorithm is an appropriate method to explore the search space of this complex problem and leads to good solutions in a reasonable amount of time.

Molecular replicators were introduced as a possible theory to explain the origin of life. Since their proposal they have been extensively studied from a bio- chemical perspective. This work proposes a taxonomy for the main properties of replicators that are important for building computational tools to solve complex problems as well as introduces a computer algorithm that models these entities. The simulation of this algorithm allows the observation and analysis of the behavior of replicators in light of the properties introduced. A number of experiments are performed to show that the proposed taxonomy of properties can be observed by simulating the algorithm introduced.

Over the last decades, Particle Filter also known as the Sampling Importance Resampling algorithm has successfully been applied to solve different problems in Engineering, e.g., trajectory tracking, non-linear estimation, and many others. Basically, the Particle Filter algorithm consists of a population of particles, which are sampled to estimate a posterior probability distribution. Unfortunately, in some cases the algorithm suffers from particle degeneracy, in which most particles converge prematurely to local minima due a loss of diversity of the population, and therefore do not contribute to estimation of the true probability distribution. In this paper, in order to tackle this drawback and to improve the performance of the standard Particle Filter we propose a modification to the algorithm by inserting a sampling mechanism inspired by Differential Evolution. Simulation results of the enhanced hybrid version are presented and compared with the standard Particle Filter algorithm and show the suitability of the proposed approach.

In this paper, we propose a novel normal parameter reduction algorithm of soft sets based on the oriented-parameter sum, which can be carried out without parameter important degree and decision partition. We present some new related definitions and proved theorems of normal parameter reduction. The comparison result on a Boolean-valued dataset shows that, the proposed algorithm involves relatively less computation and is easier to implement and understand as compared with the soft set-based algorithm of normal parameter reduction.

Data Envelopment Analysis (DEA) is one of the popular approaches of decision analysis. Parametric Settings for DEA is one of the essential steps for the decision making. This research proposes the method to apply Cognitive Pairwise Comparison (CPC) to the determination of the parametric settings in DEA. The usability and applicability of the enhanced DEA are demonstrated in a resource allocation problem on the basis of quality-cost balance.

We consider an extension of the Resource-Constrained Project Scheduling Problem (RCPSP) to multi-level (or multi-mode) activities. Each activity must be allocated exactly one unit of each required resource and the resource unit may be used at any of its specified levels. The processing time of an activity is given by the maximum of the durations that would result from a specific allocation of resources. The objective is to find the optimal solution that minimizes the overall project cost which includes a penalty for tardiness beyond the specified delivery date as well as a bonus for early delivery. We give some of the most important solution details and we report on the preliminary results obtained. The implementation was designed using the C# language.

In this paper, a strict authentication watermarking scheme based on multi-modal biometric images and near sets was designed and introduced. The proposed scheme has a number of stages including feature enrolment for extracting the human facial features. Three human facial features which are nose length, nose width and distance between eyes balls are extracted. The near sets approach is adapted to choose the best feature among the considered features. The watermark is generated from hashing the extracted facial features that then encrypted using Advanced Encryption Standard (AES) technique and embedding the encrypted value into the human fingerprint image in order to confirm the integrity of respective biometric data. The experimental result shows that the proposed scheme guarantees the security assurance.

This paper proposes an ant colony optimization (ACO) algorithm to deal with fuzzy document clustering problems. A specialized glossary and a thesaurus are used in order to extract features of the documents and to obtain a languageindependent vector representation that can be used to measure similarities between documents written in different languages. The pheromone trails obtained in the ACO process are used to determine membership values in a fuzzy clustering. To illustrate the behavior of the algorithm, it was applied to a corpus of bilingual documents in different areas of economic and management.

The support vector machine (SVM) classifier is currently one of the most powerful techniques for solving binary classification problems. To further increase the accuracy of an individual SVM we use an ensemble of SVMs, composed of classifiers that are as accurate and divergent as possible.We investigate the usefulness of SVM ensembles in which the classifiers differ among themselves in both the feature set and the SVM parameter value they use, which might increase the diversity among the classifiers and therefore the ensemble accuracy.We propose a novel method for building an accurate SVM ensemble. First we perform complementary feature selection methods to generate a set of feature subsets, and then for each feature subset we build a SVM classifier which uses tuned SVM parameters. The experiments show that this method achieved a higher estimated prediction accuracy in comparison to well-established approaches for building SVM ensembles, namely using a Genetic Algorithm based search to vary the classifier feature sets and using a predefined set of SVM parameter values to vary the classifier parameters.We work in a context of real-world industrial machine fault diagnosis, using 2000 examples of vibrational signals obtained from operating faulty motor pumps installed on oil platforms.

Sometimes, the soil foundation is inadequate for constructions purpose (soft-soils). In these cases there is need to improve its mechanical and physical properties. For this purpose, there are several geotechnical techniques where Jet Grouting (JG) is highlighted. In many geotechnical structures, advance design incorporates the ultimate limit state (ULS) and the serviceability limit state (SLS) design criteria, for which uniaxial compressive strength and deformability properties of the improved soils are needed. In this paper, three Data Mining models, i.e. Artificial Neural Networks (ANN), Support Vector Machines (SVM) and Functional Networks (FN), were used to estimate the tangent elastic Young modulus at 50% of the maximum stress applied (

E

tg

50%

) of JG laboratory formulations over time. A sensitivity analysis procedure was also applied in order to understand the influence of each parameter in

E

tg

50%

estimation. It is shown that the data driven model is able to learn the complex relationship between

E

tg

50%

and its contributing factors. The obtained results, namely the relative importance of each parameter, were compared with the predictive models of elastic Young modulus at very small strain (

E

0

) as well as the uniaxial compressive strength (

Q

u

). The obtained results can help to understand the behavior of soil-cement mixtures over time and reduce the costs with laboratory formulations.

This paper introduces a hybrid scheme that combines the advantages of deep belief network and support vector machine. An application of intrusion detection imaging has been chosen and hybridization scheme have been applied to see their ability and accuracy to classify the intrusion into two outcomes: normal or attack, and the attacks fall into four classes; R2L, DoS, U2R, and Probing. First, we utilize deep belief network to reduct the dimensionality of the feature sets. This is followed by a support vector machine to classify the intrusion into five outcome; Normal, R2L, DoS, U2R, and Probing. To evaluate the performance of our approach, we present tests on NSL-KDD dataset and show that the overall accuracy offered by the employed approach is high.

In this paper, we present a collaborative multi-agent based system for mining association rules from distributed databases. The proposed model is based on cooperative agents and is compliant to the Foundation for Intelligent Physical Agents standard. This model combines different types of technologies, namely the association rules as a data mining technique and the multi-agent systems to build a model that can operate on distributed databases rather than working on a centralized database only. The autonomous and the social abilities of the model agents provided the ability to operate cooperatively with each other and with other different external agents, thus offering a generic platform and a basic infrastructure that can deal with other data mining techniques. The platform has been compared with the traditional association rules algorithms and has proved to be more efficient and more scalable.

Quantum Evolutionary Algorithm (QEA) is a novel optimization algorithm proposed for class of combinatorial optimization problems.While Fractal Image Compression problem is considered as a combinatorial problem, QEA is not widely used in this problem yet. Using the spatial correlation between the neighbouring blocks, this paper proposes a novel initialization method for QEA. In the proposed method the information gathered from the previous searches for the neighbour blocks is used in the initialization step of search process of range blocks. Then QEA starts searching the search space to find the best matching domain block. The proposed algorithmis tested on several images for several dimensions and the experimental results shows better performance for the proposed algorithm than QEA and GA. In comparison with the full search algorithm, the proposed algorithm reaches comparable results with much less computational complexity.

The sound of road vehicles plays a major role in providing quiet and comfortable rides. Automotive companies have invested a great deal over the last few decades to achieve this goal and attract customers. Engine noise has become one of the major sources of passenger car noise today and the demand for accurate prediction models is high. The purpose of this paper is to develop a novel noise prediction model in vehicles using a Pass-by noise test based on Artificial Neural Networks at high frequencies. The artificial neural network used in the experiments was the Generalized Gaussian Radial Basis Function Neural Network (GRBFNN). This type of RBF can reproduce different RBFs by updating a real

τ

parameter and allowing different shapes of RBFs in the same Neural Network. At low frequencies the system behaves linearly and therefore the proposed method improves the accuracy of the system in frequencies over 2.5 kH, obtaining a Mean Squared Error (MSE) of 0.018 ±3×10

− 4

, enough for our noise prediction aim.

Numerous research efforts have tried to mimic the capabilities of living organisms in performing self-monitoring and self-repairing denoted as self-x features to achieve robust and dependable systems. In sensor systems applications, self-x features carry the promise to deliver properties requested by standards organizations, e.g., the NAMUR[1], such as improved flexibility, better accuracy and reduced vulnerability to deviations and drift caused by manufacturing and the environmental changes. In this paper, the concept of self-x properties implemented on an Anisotropic Magnetoresistive AMR sensor system is investigated as a first case study to be carried on in MEMS implementations. The degradation of AMR sensor can occur when the sensor is exposed to the strong magnetic field shown by weak sensitivity of sensor and inaccurate measurement output. The self-x properties are required to monitor and recover the sensor performance by employing the compensating and flipping coils. The experimental result shows the recovering of sensor performance in terms of classification accuracy for vehicle recognition application by implementing the self-x features.

In modern container terminals, efficiently managing the transit of the containers becomes more and more of a challenge. Due to the progressive evolution of container transport, traffic management within container ports is still an evolving problem. To provide adequate strategy for the increased traffic, ports must either expand facilities or improve efficiency of operations. In investigating ways in which ports can improve efficiency, this paper proposes a Markov Decision Process (MDP) for loading and unloading operations within a container terminal. The proposed methodology allows an easy modeling for optimizing complex sequences of decisions that are to be undertaken at each time. The goal is to minimize the total waiting time of quay cranes and vehicles, which are allocated to service a containership. In this paper, reinforcement learning, which consists of solving learning problems by studying the system through mathematical analysis or computational experiments, is considered to be the most adequate approach.

In this paper a Generalized Least Square estimator for the simultaneous estimation of O-D matrix and equilibrium traffic assignment model parameters is presented. The problem is formulated as fixed-point model (equilibrium programming) assuming the congested network case. In the optimization step the variability of both O-D demand vector and the matrix of link choice probabilities is considered. We assume as input information a set of observable network data, such as link traffic counts and travel time, as well starting estimates of both O-D matrix and models parameters. Along the paper, the theoretical aspects of the proposed estimator, the solution algorithm as well as the results of numerical applications are discussed.

Transport companies in many cases have to evaluate their competitiveness, comparing it with that of their competitors. Usually this assessment is performed through one or more indices representing facility performances, derived from a set of indicators relevant to problem representation. If the aim is to estimate the user evaluation for the service offered by a facility, the development of a synthetic index can be difficult since user’s choice is often characterized by significant uncertainties and it is not always governed by certain rules and rational behaviour, so that it could not be easily and explicitly represented by traditional mathematical techniques and models. Such uncertainties in the relationship between indicator values and facility attractiveness can be properly defined by explicitly specifying them in an approximate way using fuzzy sets theory. In this paper an innovative approach for the classification of Transport Facilities is proposed. The method is based on a Fuzzy Inference System and may be employed both as a benchmarking/ranking procedure and as a decision support tool to evaluate future scenarios as a result of facilities remodelling.

The transferability of fuzzy models of gap-acceptance behavior between different intersections is evaluated in this paper using a method known as ROC curve analysis. The results of an application to four unsignalized intersections indicate that, even if transferred models generally perform adequately, intersection geometric and traffic characteristics may be very important in determining the capability of a model developed in a given context to reproduce gap-acceptance behavior observed in other contexts.

Quantum-dot cellular automata (QCA) are proposed as one of the foremost candidates to replace the complementary metal-oxide semiconductor (CMOS) technology. The majority gate and the inverter gate together make a universal set of Boolean primitives in QCA technology. Reducing the number of required primitives to implement a given Boolean function is an important step in designing QCA logic circuits. Previous research has shown how to use genetic algorithms to minimize the number of majority gates implementing a given Boolean function with one output. In this paper we show how to minimize Boolean functions with an arbitrary number of outputs. Simulation results for the circuits with three, four and five outputs show our method on the average results in 25.41, 28.82, 30.89 percentage decrease in the number of required gates in comparison with optimizing each output independently.

The design of an optimized logic circuit that implements a desired Boolean function is of interest. Optimization can be performed in terms of different objectives. They include optimizing the number of gates, the number of levels, the number of transistors of the circuit, etc. In this paper, we describe an approach using genetic programming to optimize a given Boolean function concerning the above mentioned objectives. Instead of commonly used set of gates, i.e. {AND, OR, NOT, XOR}, we use the universal NAND gates which lead to a faster and more compact circuit. The traditional gate minimization techniques produce simplified expressions in the two standard forms: sum of products (SOP) or product of sums (POS). The SOP form can be transformed to a NAND expression by a routine, but the transformation does not lead to optimized circuit; neither in terms of the number of gates, nor the number of levels. Experimental results show our approach produces better results compared to transforming the SOP form to the NAND expression, with respect to the number of gates, levels and transistors of the circuit.

This paper presents an augmented Lagrangian algorithmto solve continuous constrained global optimization problems. The algorithm approximately solves a sequence of bound constrained subproblems whose objective function penalizes equality and inequality constraints violation and depends on the Lagrange multiplier vectors and a penalty parameter. Each subproblem is solved by a population-based method that uses an electromagnetism-like mechanism to move points towards optimality. Benchmark problems are solved in a performance evaluation of the proposed augmented Lagrangian methodology.A comparison with a well-known technique is also reported.

In this work, it is described a gait multiobjective optimization system that allows to obtain fast but stable robot quadruped crawl gaits. We combine bioinspired Central Patterns Generators (CPGs) and Genetic Algorithms (GA). A motion architecture based on CPGs oscillators is used to model the locomotion of the robot dog and a GA is used to search parameterizations of the CPGs parameters which minimize the body vibration, maximize the velocity and maximize the wide stability margin. In this problem, there are several conflicting objectives that leads to a multiobjective formulation that is solved using the Weighted Tchebycheff scalarization method. Several experimental results show the effectiveness of this proposed approach.