Tae-Chon Ahn

Parameter estimation of fuzzy controller and its application to inverted pendulum

Abstract In this paper, a new approach to estimate scaling factors of the fuzzy PID controller is presented. The performance of the fuzzy PID controller is sensitive to the variety of scaling factors. The design procedure dwells on the use of evolutionary computing (more specifically, a genetic algorithm) and estimation algorithm. The tuning of the scaling factors of the fuzzy PID controller is essential to the entire optimization process. And then we estimate scaling factors of the fuzzy PID controller by means of three types of estimation algorithms such as HCM (Hard C-Means) clustering-based regression polynomial, neuro-fuzzy networks, and regression polynomials. Numerical studies are presented in detail along with a detailed comparative analysis.

Self-organizing neural networks with fuzzy polynomial neurons

Abstract We propose a new category of neurofuzzy networks—self-organizing neural networks (SONNs) with fuzzy polynomial neurons (FPNs). For these networks we discuss a comprehensive design methodology. SONN dwells on the ideas of fuzzy rule-based computing and neural networks thus emerging as a sound construct of computational intelligence. SONN is a flexible neural architecture whose structure is not fixed in advance but becomes developed through a growth (generation of layer) process. In particular, the number of layers of the SONN is not fixed in advance unlike in the case of conventional neural networks, but is generated in a dynamic way. Simulation studies involve a series of synthetic as well as experimental data used across various neurofuzzy systems. A comparative analysis is included as well.

Obstacle Classification by a Line-Crawling Robot: A Rough Neurocomputing Approach

This article considers a rough neurocomputing approach to the design of the classify layer of a Brooks architecture for a robot control system. In the case of the line-crawling robot (LCR) described in this article, rough neurocomputing is used to classify sometimes noisy signals from sensors. The LCR is a robot designed to crawl along high-voltage transmission lines where noisy sensor signals are common because of the electromagnetic field surrounding conductors. In rough neurocomputing, training a network of neurons is defined by algorithms for adjusting parameters in the approximation space of each neuron. Learning in a rough neural network is defined relative to local parameter adjustments. Input to a sensor signal classifier is in the form of clusters of similar sensor signal values. This article gives a very brief description of a LCR that has been developed over the past three years as part of a Manitoba Hydro research project. This robot is useful in solving maintenance problems in power systems. A description of the basic features of the LCR control system and basic architecture of a rough neurocomputing system for robot navigation are given. A sample LCR sensor signal classification experiment is also given.

The design methodology of radial basis function neural networks based on fuzzy K-nearest neighbors approach

Various approaches to partitioning of high-dimensional input space have been studied with the intent of developing homogeneous clusters formed over input and output spaces of variables encountered in system modeling. In this study, we propose a new design methodology of a fuzzy model where the input space is partitioned by making use of some classification algorithm, especially, fuzzy K-nearest neighbors (K-NN) classifier being guided by some auxiliary information granules formed in the output space. This classifier being regarded in the context of this design as a supervision mechanism is used to capture the distribution of data over the output space. This type of supervision is beneficial when developing the structure in the input space. It is demonstrated that data involved in a partition constructed by the fuzzy K-NN method realized in the input space show a high level of homogeneity with regard to the data present in the output space. This enhances the performance of the fuzzy rule-based model whose premises in the rules involve partitions formed by the fuzzy K-NN. The design is illustrated with the aid of numeric examples that also provide a detailed insight into the performance of the fuzzy models and quantify several crucial design issues.

A design of granular fuzzy classifier

In this paper, we propose a new design methodology of granular fuzzy classifiers based on a concept of information granularity and information granules. The classifier uses the mechanism of information granulation with the aid of which the entire input space is split into a collection of subspaces. When designing the proposed fuzzy classifier, these information granules are constructed in a way they are made reflective of the geometry of patterns belonging to individual classes. Although the elements involved in the generated information granules (clusters) seem to be homogeneous with respect to the distribution of patterns in the input (feature) space, they still could exhibit a significant level of heterogeneity when it comes to the class distribution within the individual clusters. To build an efficient classifier, we improve the class homogeneity of the originally constructed information granules (by adjusting the prototypes of the clusters) and use a weighting scheme as an aggregation mechanism.

Design of neuro-fuzzy controller on DSP for real-time control of induction motors

This paper deals with the DSP implementation of the high performance induction motor drive that presented on the viewpoint of the design and experiment. The speed control system for the induction motor drive is based on the ANFIS (adaptive network-based fuzzy inference system) controller, that is, a sophisticated neuro-fuzzy controller. This ANFIS controller acts as a feed forward controller that provides the plant with the proper control input and accomplish error backpropagation algorithm through the network. In this paper, the DSP (TMS320F240) has been used to perform the high-speed calculation of the space vector PWM and to build the ANFIS control algorithm. It is confirmed that proposed algorithm provides the more improved control performance for the conventional V/F controller and vector controller. The proposed ANFIS algorithm and DSP technique can be applied to the precise speed control of the induction motor drive system or the field of power electronics.

The refinement of models with the aid of the fuzzy k-nearest neighbors approach

In this paper, we propose a new design methodology that supports the development of hybrid incremental models. These models result through an iterative process in which a parametric model and a nonparametric model are combined so that their underlying and complementary functionalities become fully exploited. The parametric component of the hybrid model captures some global relationships between the input variables and the output variable. The nonparametric model focuses on capturing local input-output relationships and thus augments the behavior of the model being formed at the global level. In the underlying design, we consider linear and quadratic regression to be a parametric model, whereas a fuzzy k-nearest neighbors model serves as the nonparametric counterpart of the overall model. Numeric results come from experiments that were carried out on some low-dimensional synthetic data sets and several machine learning data sets from the University of California-Irvine Machine Learning Repository.

Fuzzy linear regression based on Polynomial Neural Networks

Highlight? The new estimation approach to determine parameters of fuzzy regression is proposed. ? A new design methodology of fuzzy is based on Polynomial Neural Networks. ? Fuzzy numbers for fuzzy regression are formed by Particle Swarm Optimization. In this study, we introduce an estimation approach to determine the parameters of the fuzzy linear regression model. The analytical solution to estimate the values of the parameters has been studied. The issue of negative spreads of fuzzy linear regression makes the problem to be NP complete. To deal with this problem, an iterative refinement of the model parameters based on the gradient decent optimization has been introduced.In the proposed approach, we use a hierarchical structure which is composed of dynamically accumulated simple nodes based on Polynomial Neural Networks the structure of which is very flexible.In this study, we proposed a new methodology of fuzzy linear regression based on the design method of Polynomial Neural Networks. Polynomial Neural Networks divide the complicated analytical approach to estimate the parameters of fuzzy linear regression into several simple analytic approaches.The fuzzy linear regression is implemented by Polynomial Neural Networks with fuzzy numbers which are formed by exploiting clustering and Particle Swarm Optimization. It is shown that the design strategy produces a model exhibiting sound performance.

Design of Fuzzy Set-based Polynomial Neural Networks involving Information Granules with the aid of multi-population Genetic Algorithms

In this paper, we propose a new design methodology of fuzzy-neural networks - fuzzy set-based polynomial neural networks (FSPNN) with symbolic genetic algorithms. Fuzzy set-based polynomial neural networks (FSPNN) are based on a fuzzy set-based polynomial neuron (FSPN) whose fuzzy rules include the information granules (about the real system) obtained through information granulation. The information granules are capable of show the specific characteristic of the system. We have developed a design methodology (genetic optimization using symbolic genetic algorithms) to find the optimal structure for fuzzy-neural networks that expanded from group method of data handling (GMDH). It is the number of input variables, the order of the polynomial, the number of membership functions, and a collection of the specific subset of input variables that are the parameters of FSPNN fixed by aid of symbolic genetic optimization that has search capability to find the optimal solution on the solution space. The augmented and genetically developed FPNN (gFPNN) results in a structurally optimized structure and comes with a higher level of flexibility in comparison to the one we encounter in the conventional FPNNs. The GA-based design procedure being applied at each layer of FPNN leads to the selection of the most suitable nodes (or FSPNs) available within the FPNN. Symbolic genetic algorithms are capable of reducing the solution space more than conventional genetic algorithms with binary genetype chromosomes. The performance of genetically optimized FSPNN (gFSPNN) with aid of symbolic genetic algorithms is quantified through experimentation where we use a number of modeling benchmarks data which are already experimented with in fuzzy or neurofuzzy modeling

IG-based genetically optimized fuzzy polynomial neural networks with fuzzy set-based polynomial neurons

In this study, we introduce and investigate a new topology of fuzzy-neural networks-fuzzy polynomial neural networks (FPNN) that is based on a genetically optimized multiplayer perceptron with fuzzy set-based polynomial neurons (FSPNs). We also develop a comprehensive design methodology involving mechanisms of genetic optimization and information granulation. In the sequel, the genetically optimized FPNN (gFPNN) is formed with the use of fuzzy set-based polynomial neurons (FSPNs) composed of fuzzy set-based rules through the process of information granulation. This granulation is realized with the aid of the C-means clustering (C-Means). The design procedure applied in the construction of each layer of an FPNN deals with its structural optimization involving the selection of the most suitable nodes (or FSPNs) with specific local characteristics (such as the number of input variable, the order of the polynomial, the number of membership functions, and a collection of specific subset of input variables) and address main aspects of parametric optimization. Along this line, two general optimization mechanisms are explored. The structural optimization is realized via genetic algorithms (GAs) and HCM method whereas in case of the parametric optimization we proceed with a standard least square estimation (learning). Through the consecutive process of structural and parametric optimization, a flexible neural network is generated in a dynamic fashion. The performance of the designed networks is quantified through experimentation where we use two modeling benchmarks already commonly utilized within the area of fuzzy or neurofuzzy modeling.