Jyh-Shing Roger Jang

Neuro-fuzzy modeling and control

Fundamental and advanced developments in neuro-fuzzy synergisms for modeling and control are reviewed. The essential part of neuro-fuzzy synergisms comes from a common framework called adaptive networks, which unifies both neural networks and fuzzy models. The fuzzy models under the framework of adaptive networks is called adaptive-network-based fuzzy inference system (ANFIS), which possess certain advantages over neural networks. We introduce the design methods for ANFIS in both modeling and control applications. Current problems and future directions for neuro-fuzzy approaches are also addressed. >

Self-learning fuzzy controllers based on temporal backpropagation

A generalized control strategy that enhances fuzzy controllers with self-learning capability for achieving prescribed control objectives in a near-optimal manner is presented. This methodology, termed temporal backpropagation, is model-sensitive in the sense that it can deal with plants that can be represented in a piecewise-differentiable format, such as difference equations, neural networks, GMDH structures, and fuzzy models. Regardless of the numbers of inputs and outputs of the plants under consideration, the proposed approach can either refine the fuzzy if-then rules of human experts or automatically derive the fuzzy if-then rules if human experts are not available. The inverted pendulum system is employed as a testbed to demonstrate the effectiveness of the proposed control scheme and the robustness of the acquired fuzzy controller.

Functional equivalence between radial basis function networks and fuzzy inference systems

It is shown that, under some minor restrictions, the functional behavior of radial basis function networks (RBFNs) and that of fuzzy inference systems are actually equivalent. This functional equivalence makes it possible to apply what has been discovered (learning rule, representational power, etc.) for one of the models to the other, and vice versa. It is of interest to observe that two models stemming from different origins turn out to be functionally equivalent.

Input selection for ANFIS learning

We present a quick and straightfoward way of input selection for neuro-fuzzy modeling using adaptive neuro-fuzzy inference systems (ANFIS). The method is tested on two real-world problems: the nonlinear regression problem of automobile MPG (miles per gallon) prediction, and the nonlinear system identification using the Box and Jenkins gas furnace data.

Fuzzy controller design without domain experts

The control of nonlinear systems through a self-learning mechanism that can derive the membership functions of the rules used by a fuzzy controller is considered. Without resorting to domain experts, a fuzzy controller has to be constructed that can perform the control task of a regulator problem. The approach is based on the adaptive network, a flexible building block that can be used to implement fuzzy controllers as well as the plants under consideration. The learning rule of adaptive networks can force the plant state to approach a desired state on a time step by time step basis. The proposed approach was used to build a fuzzy controller for balancing an inverted pendulum system. It is shown that only four fuzzy if-then rules are necessary to perform the control task. The controller was quite tolerant to dealing with initial conditions that deviated significantly from the origin. The inverted pendulum system was used to test the proposed control scheme. The simulation results demonstrated its feasibility and robustness.<<ETX>>

Structure determination in fuzzy modeling: a fuzzy CART approach

This paper presents an innovative approach to the structure determination problem in fuzzy modeling. By using the well-known CART (classification and regression tree) algorithm as a quick preprocess, the proposed method can roughly estimate the structure (numbers of membership functions and number of fuzzy rules, etc.) of a fuzzy inference system; the parameter identification is then carried out by the hybrid learning scheme developed in our previous work. Moreover, the identified fuzzy inference system has the property that the total of firing strengths is always equal to one; this speeds up learning processes and reduces round-off errors.<<ETX>>

Predicting chaotic time series with fuzzy if-then rules

The authors continue work on a previously proposed ANFIS (adaptive-network-based fuzzy inference system) architecture, with emphasis on the applications to time series prediction. They show how to model the Mackey-Glass chaotic time series with 16 fuzzy if-then rules. The performance obtained outperforms various standard statistical approaches and artificial neural network modeling methods reported in the literature. Other potential applications of ANFIS are also suggested.<<ETX>>

Coactive neural fuzzy modeling

We discuss the neuro-fuzzy modeling and learning mechanisms of CANFIS (coactive neuro-fuzzy inference system) wherein both neural networks and fuzzy systems play active roles together in an effort to reach a specific goal. Their mutual dependence presents unexpected learning capabilities. CANFIS has extended the basic ideas of its predecessor ANFIS (adaptive network-based fuzzy inference system): the ANFIS concept has been extended to any number of input-output pairs. In addition, CANFIS yields advantages from nonlinear fuzzy rules. In light of some model-related limitations, this paper serves to highlight both neuro-fuzzy learning capacities and practical obstacles encountered in performing neuro-fuzzy modeling.