Hsuan-Ming Feng

Self-generation RBFNs using evolutional PSO learning

Abstract In this paper, a powerful evolutional particle swarm optimization (PSO) learning algorithm is developed that self-generates radial basis function neural networks (RBFNs) to deal with three nonlinear problems. In general, the simple but efficient PSO learning algorithm is adapted for solving complex and global optimization problems. With computer simulation, this paper illustrates, in detail, how the PSO algorithm can automatically tune the centers and spreads of each radial basis function, and the connection weights. Furthermore, in parallel with the free RBFNs parameters, the number of radial basis functions of the constructed RBFNs can be automatically minimized by choosing a special fitness function which takes this factor into account. Therefore, the proposed PSO allows high accuracy within a short training time when determining RBFNs with small numbers of radial basis functions. Simulation results demonstrate the proposed RBFNs’ efficiency to stabilize an inverted pendulum, approximate a nonlinear function and approximate a discrete dynamic system.

Evolutionary fuzzy particle swarm optimization vector quantization learning scheme in image compression

Abstract This article develops an evolutional fuzzy particle swarm optimization (FPSO) learning algorithm to self extract the near optimum codebook of vector quantization (VQ) for carrying on image compression. The fuzzy particle swarm optimization vector quantization (FPSOVQ) learning schemes, combined advantages of the adaptive fuzzy inference method (FIM), the simple VQ concept and the efficient particle swarm optimization (PSO), are considered at the same time to automatically create near optimum codebook to achieve the application of image compression. The FIM is known as a soft decision to measure the relational grade for a given sequence. In our research, the FIM is applied to determine the similar grade between the codebook and the original image patterns. In spite of popular usage of Linde–Buzo–Grey (LBG) algorithm, the powerful evolutional PSO learning algorithm is taken to optimize the fuzzy inference system, which is used to extract appropriate codebooks for compressing several input testing grey-level images. The proposed FPSOVQ learning scheme compared with LBG based VQ learning method is presented to demonstrate its great result in several real image compression examples.

Particle swarm optimization learning fuzzy systems design

A particle swarm optimization (PSO) learning algorithm is proposed in our research to generate fuzzy systems for balancing the car-pole system and approximating a nonlinear function. Trust to the devoted feature of PSO, i.e. simple implementation, fast convergence and small computational load, this paper illustrates the perfect PSO algorithm in detail with computer simulation to automatically tune some adjustable parameters of fuzzy systems. Computer simulation results on two nonlinear problems are derived to demonstrate the powerful PSO learning algorithm.

Short Communication: Intelligent omni-directional vision-based mobile robot fuzzy systems design and implementation

An evolutional particle swarm optimization (PSO)-learning algorithm is proposed to automatically generate fuzzy decision rules. Due to the development of the fuzzy rule-based system, it actually regulates the omni-directional vision-based mobile robot for obstacle avoidance and desired target approximation as soon as possible. In the proposed image processing algorithm, an image direct transformation method is applied to convert the omni-directional scene into panoramic normal-view. Thus, the objects positions of obstacle and target are detected by the proposed color image segmentation. Human knowledge-based fuzzy systems demonstrate their well adaptability for nonlinear and time-variant features of the mobile robot to actually approach the desired location whatever it is surrounded in a known or unknown environment. In software simulations, the omni-directional mobile robot can move toward desired targets from different initial positions and various block sizes. In hardware implementations, the fuzzy control system embedded in actual mobile robot platform is used to real-time manipulate the omni-directional wheels through the motor drivers by the captured image positions of the obstacle and target. The selected fuzzy rules are efficient to control the direction and speed of omni-directional wheels to achieve the desired targets.

Hybrid intelligent vision-based car-like vehicle backing systems design

This paper is integrated with discrete wavelet transformation (DWT), self-organizing map (SOM) neural network and fuzzy logic control methods to approach the intelligent vision-based car-like vehicle backing system. All the presented procedures are implemented as software simulations and are synthesized as hardware to describe their superiorities in our experimental platform. In a word, this study has been implemented as a car-like vehicle consists of a field programmable gate array (FPGA) chip, a CMOS image sensor, a microprocessor, a flash memory module, and servo motors. From our experimental results, this paper not only illustrates the results with computer simulation, but also demonstrates the practical car parking achievements in a real environment.

SELF-GENERATION FUZZY MODELING SYSTEMS THROUGH HIERARCHICAL RECURSIVE-BASED PARTICLE SWARM OPTIMIZATION

ABSTRACT In this article, fuzzy modeling systems are automatically developed by Hierarchical Recursive-Based Particle Swarm Optimization (HRPSO). HRPSO, which contains Fuzzy C-Mean (FCM) clustering, Particle Swarm Optimization (PSO), and recursive least-squares technology, self constructs adjustable parameters for approximating a nonlinear function and a discrete dynamic system. In general, the heuristic PSO is an evolutional computing technology when solving complex and global problems. However, the necessary training time is unsuitable for large population sizes and many adjustable parameters. To quickly approximate the actual output of nonlinear functions, the input-output training data is initially clustered by an FCM algorithm. From there favorable features are extracted from the training data and some fuzzy structures with fewer adjustable parameters will be collected as the initial population of the PSO. The FCM procedure is used to directly extract necessary small populations of PSO from training samples. After that, the recursive-based PSO is proposed to tune some adjustable parameters to quickly construct the desired fuzzy modeling system. Therefore, the proposed HRPSO determines fuzzy modeling systems with a small number of fuzzy rules to approach high accuracy within a short training time. Simulation results demonstrate the efficiency of our fuzzy model systems to solve two nonlinear problems.

ADAPTIVE HYPER-FUZZY PARTITION PARTICLE SWARM OPTIMIZATION CLUSTERING ALGORITHM

This article presents an adaptive hyper-fuzzy partition particle swarm optimization clustering algorithm to optimally classify different geometrical structure data sets into correct groups. In this architecture, we use a novel hyper-fuzzy partition metric to improve the traditional common-used Euclidean norm metric clustering method. Since one fuzzy rule describes one pattern feature and implies the detection of one cluster center, it is encouraged to decrease the number of fuzzy rules with the hyper-fuzzy partition metric. According to the adaptive particle swarm optimization, it is very suitable to manage the clustering task for a complex, irregular, and high dimensional data set. To demonstrate the robustness of the proposed adaptive hyper-fuzzy partition particle swarm optimization clustering algorithms, various clustering simulations are experimentally compared with K-means and fuzzy c-means learning methods.

A self-tuning fuzzy control system design

A self-tuning fuzzy control system is proposed such that the controlled system has the desired output without knowing the mathematical model of the system. In this control structure, a parameter tuning algorithm based on the technology of reinforcement learning and decision making is constructed to enable it to tune the consequent parameters of the fuzzy controller such that the fuzzy controller has a self-tuning ability. In this paper, a state evaluator is considered to play the role of a critical element to evaluate the current state of the controlled system. A functional-type evaluator is used to produce a scalar value, which is provided to a parameter modifier to tune the adjustable parameters of the fuzzy controller. A decision making mechanism works as a director to select an appropriate parameter and get a better action such that the controlled system has better performance. The goal of the parameter tuning algorithm is to maximize the evaluation value of the current state such that the control objective can be attained. Finally, the inverted pendulum control problem is used to illustrate the effectiveness of the proposed control system structure.

DESIGN OF FUZZY SYSTEMS WITH FEWER RULES

This article presents an innovative method for designing fuzzy systems composed of fewer fuzzy rules. The conventional approach to fuzzy-system design usually assumes that there exists no correlation among input variables, therefore, grid-type fuzzy partitions are widely adopted. However, in many cases, it is likely that input variables are highly correlated with one another. To avoid the problem of growth of partitioned grids in some complex system, we used an aggregation of hyperrectangulars with different size and different positions to approximate fuzzy partitions that are arbitrarily shaped. The corresponding parameters defining these hyperrectangulars are selected by using genetic algorithms. Furthermore, the number of fuzzy rules of the constructed system can be automatically minimized by choosing a special fitness function that takes this factor into account. Finally, an inverted pendulum control and nonlinear modeling problems are utilized to illustrate the effectiveness of the proposed method.

A on-line rule tuning grey prediction fuzzy control system design

An on-line rule tuning grey prediction fuzzy control system is presented, which contains the advantages of grey prediction, fuzzy theory and the on-line tuning algorithm. The on-line rule tuning grey prediction fuzzy control system structure is constructed so that the rise time and the overshoot of the controlled system can be maintained simultaneously.