Zhenlei Wang

Chaos-Genetic Algorithm for Multiobjective Optimization

Chaos-genetic algorithm (CGA) combining local chaotic search and nondominated sorting genetic algorithm for multiobjective optimization is proposed. The method is composed of two stages. The wide search with nondominated sorting genetic algorithm (NSGA-II) is performed at the first searching stage, then the local search with chaotic mutation is performed at the second stage. Moreover, we cancel the limitation of the number of the elitism at each generation and improve the original clustering method. We apply the coverage measure and spread measure to evaluate the performance of the two methods, and obtain more satisfactory results with CGA than that with NSGA-II

An Expert System for Real-time Fault Diagnosis and Its Application in PTA Process

Increasing complexity and automation of the chemical process industries requires more reliable and efficient real-time fault diagnosis systems. Here, a real-time expert system based on wavelet transform and fuzzy ART neural network is introduced for fault diagnosis, providing fault prediction to help operators before abnormal situations occur. Data preprocessing, knowledge base structure, representation of knowledge, generalized inference engine and graphic user interface are technically considered. Industrial applications in pure terephthalic acid (PTA) process indicate that the real-time expert system diagnoses abnormal events efficiently and promptly and it has many specialties such as friendly interface, easy to train and maintain and also reliable under changing process conditions

Multiple model self-tuning control for a class of nonlinear systems

This study develops a novel nonlinear multiple model self-tuning control method for a class of nonlinear discrete-time systems. An increment system model and a modified robust adaptive law are proposed to expand the application range, thus eliminating the assumption that either the nonlinear term of the nonlinear system or its differential term is global-bounded. The nonlinear self-tuning control method can address the situation wherein the nonlinear system is not subject to a globally uniformly asymptotically stable zero dynamics by incorporating the pole-placement scheme. A novel, nonlinear control structure based on this scheme is presented to improve control precision. Stability and convergence can be confirmed when the proposed multiple model self-tuning control method is applied. Furthermore, simulation results demonstrate the effectiveness of the proposed method.

Multiobjective evolutionary algorithm based on the Pareto Archive and individual migration

A multiobjective evolutionary algorithm based on the parallel evolution of multiple single objective populations and Pareto archive population is proposed. For each single objective population, single objective evolutionary algorithm is applied to optimize separately each of multiobjective functions, where individuals generated by tournament selection from the union of single objective and Pareto archive population form the single objective population of next generation. At each evolving iteration, based on the concept of Pareto dominance, a finite-sized Pareto archive population is iteratively updated and trimmed by a new crowded-comparison operation. Especially, individuals in Pareto archive population also join evolutionary operations to increase the converging speed and improve quality of nondominated solutions. Simulations manifest that the proposed method can realize the search from multiple directions to obtain the nondominated solutions scattered more uniformly over the Pareto frontier with better convergence metric compared to well-known NSGA-II algorithm. Individuals migrating from Pareto archive population by tournament selection is also proved to have the advantage in improving the converging speed and converging precision.

IPSO: An immune based PSO supervised learning system for incremental learning

PSO has been proved as an effective supervised learning system in recent years, but its not an effective method for incremental learning problems. Aiming at the incremental learning target for classification, a hybrid algorithm of Particle Swarm Optimization (PSO) and Artificial Immune System (AIS) called Immune based PSO (IPSO) is presented in this paper. IPSO inherits the incremental learning ability of AIS. In IPSO, training data is presented to the algorithm one by one, and the training proceed is a one-shot incremental algorithm. Besides, the swarm does not converge to a single solution; instead, each particle is a part of the classifier, and the whole memory population is taken as the integral classifier to the problem. Compared the results of standard PSO and IPSO in several benchmark problems from the UCI data sets, we found that IPSO achieved a better classification accuracy than standard PSO in most cases. It is also competitive with some of the algorithms most commonly used for classification.

Input-to-state stability of discrete-time switched systems and switching supervisory control

In this paper, a discrete-time switched nonlinear system is proved to have an input-to-state stable (ISS) property under mode-dependent average dwell-time switching signals if each constituent subsystem is ISS. This result is then applied to stabilization of uncertain discrete-time nonlinear systems via an improved switching supervisory control scheme. The states of the closed-loop system are proved to be bounded in the presence of bounded disturbances when the candidate controllers provide ISS properties regarding the estimation errors. Simulation results are provided to illustrate the effectiveness of the proposed stabilization method.

Multiple models adaptive control based on cluster-optimization for a class of nonlinear system

For a class of nonlinear discrete time system with fast time-varying or jumping parameters, a multiple models adaptive controller (MMAC) based on cluster-optimization is proposed. Based on the input-output data, the sample data are classified into several clusters by the fuzzy kernel clustering adaptive algorithm. Then the local models can be constructed corresponding clusters by the least square method. To improve the transient response during the change of the working points, besides the distance, the directional derivative of system is computed also. It is utilized to identify the system trend of changing working point. Before the changing occurs, new weighted models are developed by the corresponding local models, indicated by the system directional derivative. Meanwhile the distance between the data and the centre of clusters are used to find the weighted coefficients. So a better approach ability can be got than that designed only by the distance. The simulation results show that the proposed controller is superior to that of the conventional multiple models controller.

Fuzzy-Based Hybrid Control for Nonlinear Multivariable System

Hybrid controller based on fuzzy logic system is designed for a class of nonlinear multivariable systems. The nonlinear systems involve plant uncertainties and external disturbances. The controller is based on a combination of the Hinfin tracking theory, fuzzy logic system and supervisor control algorithm (SCA). The robust stability of the control system is proofed in the paper. When apply the hybrid controller, all the states and signals of the closed-loop system are bounded and Hinfin tracking performance is guaranteed. Consequently, the control performance of the system is greatly improved. The hybrid controller can be applied to more general nonlinear system involving a large class of uncertainties and variations. At last the simulation result is given.