Journal of Electrical and Electronic Engineering

2015; 3(2-1): 58-61

Published online January 21, 2015 (http://www.sciencepublishinggroup.com/j/jeee)
doi: 10.11648/j.jeee.s.2015030201.23
ISSN: 2329-1613 (Print); ISSN: 2329-1605 (Online)

A review of neuro-fuzzy systems based on intelligent

control
Fatemeh Zahedi*, Zahra Zahedi
Electrical and Computer Engineering, Shiraz University, Shiraz, Iran

Email address:
Fatemeh.zahedi28@gmail.com (F. Zahedi), Zahra.zahedi28@gmail.com (Z. Zahedi)

To cite this article:

Fatemeh Zahedi , Zahra Zahedi. A Review of Neuro-Fuzzy Systems Based on Intelligent Control. Journal of Electrical and Electronic
Engineering. Special Issue: Research and Practices in Electrical and Electronic Engineering in Developing Countries.
Vol. 3, No. 2-1, 2015, pp. 58-61. doi: 10.11648/j.jeee.s.2015030201.23

Abstract: The system's ability to adapt and self-organize are two key factors when it comes to how well the system can survive
the changes to the environment and the plant they work within. Intelligent control improves these two factors in controllers.
Considering the increasing complexity of dynamic systems along with their need for feedback controls, using more complicated
controls has become necessary and intelligent control can be a suitable response to this necessity. This paper briefly describes the
structure of intelligent control and provides a review on fuzzy logic and neural networks which are some of the base methods for
intelligent control. The different aspects of these two methods are then compared together and an example of a combined method
is presented.

Keywords: Intelligent control, Neural networks, Fuzzy logic, Neuro-fuzzy

1. Introduction
In the last few decades, fuzzy logic has been identified as fuzzy control rules and improvement of those rules from
one of the most active and useful fields of research and study, experience. Some of the presented methods can have hopeful
and has found many uses in many different applications. Of its results but these methods are mainly subjective and heuristic
most important usages are non-linear control systems, time and finding the proper membership functions are done through
variant systems, and control systems whose dynamics are trial and error.
exactly known, such as servo-motor position control systems On the other hand, applying the neural network’s ability to
and robot arm control systems. learn, to fuzzy logic systems provides us with more methods
In fuzzy logic systems, decisions are made based on the that can yield hopeful results. These methods which are
inputs received in the form of linguistic variables. These referred to as neuro-fuzzy methods create a desirable
variables are grouped up in fuzzy sets and are assigned a framework for finding solutions for more complicated
degree of membership within that set which are determined by problems. If the available knowledge can be expressed in the
certain formulas that are referred to as membership functions. form of linguistic rules we can have a fuzzy logic system. And
Fuzzy logic rules are defined as preconditions in the form of if we have the necessary data and are able to learn from that
if-then rules defined based on the fuzzy sets and the response data through simulations, we can use neural networks.
of each rule is determined through fuzzy implication. Specifying the fuzzy sets, the fuzzy operators and the
Currently, there are no systematic methods for designing knowledge base are requirements for building a fuzzy logic
systems based on fuzzy logic. The easiest and the fastest system, and specifying the architecture and learning
method is to define membership functions and fuzzy rules algorithms are requirements for creating a neural network [2].
based on human-operated systems which are then tested until In the final section of this paper, the cooperative neuro-
the desired output is obtained [1]. fuzzy model is reviewed. In this method, the parameters,
Current research aims at designing systems that work on including the fuzzy rules, rule weights and fuzzy sets are
fuzzy logic and learn from their own experiences. However, determined using neural networks.
there are few articles available that discuss the creating of
 Journal of Electrical and Electronic Engineering 2015; 3(2-1): 58-61 59

2. Intelligent Control the system is capable of approaching one of these equilibrium

points subject to the global stability of the system. For
Generally, modeling most complicated processes can’t be example, according to “Fig. 2”, E(X) is an energy function
achieved by mathematical function or simple physical rules. with three equilibrium points of P1, P2 and P3. If X(0) is the
Intelligent control aims at finding a solution to this problem by initial state of the system, X(0) should converge to the closest
combining intelligent and creative characteristics of human equilibrium point.
controllers. Intelligent control performs under an environment
with lots of uncertainties and unexpected situations in a way
that keeps its integrity and consistency with that environment
and rectifies failures in the system without limits.
Intelligent controls, regardless of what structure or
configuration they have, should have the following attributes
[3]:
A. Correctness: the ability to perform a specific set of
needed operations.
B. Robustness: the ability of the system to maintain its
functions under unexpected and unusual situations that can
occur internally or externally.
C. Extensibility: the ability of the intelligent control to
support planned and unplanned upgrades (hardware or Figure 2. Energy landscape with 3 minimum points
software) without the need for re-design.
D. Reusability & compatibility: the ability to use For the defined system in the differential “(1)” and “(2)”,
sub-systems and components in different applications and the Lyapunov function (energy) would be as the “(3)”.
compatibility with new situations.
xɺ i = a i ( x i )[ bi ( x i ) − ∑ k =1 wik d k ( x k )
n
There are three basic methods for intelligent control two of (1)
which are fuzzy logic and neural networks.
wɺ i = wij + G ( wij , xi , x j ) (2)
3. Neural network xi
∑ ∫ ∑
n n
V ( x) = − bi (ε i )d i′ (ε i )dε i + 1
2 w jk d j ( x j )d k ( x k ) (3)
Control engineers consider neural networks to be large i =1 0 j , k =1

scale, non-linear dynamic systems that are defined within a

first order differential equation. Neural networks are in fact 4. Fuzzy Logic
new structures for information processing systems that consist
of numerous linked processing elements. The links that 4.1. Defining some Fuzzy Operations
connect these processing elements are called interconnections.
“Fig. 1” is an example of a single processing element. Each We start out by defining some fuzzy operations. If A and B
processing element has one or more inputs and a single output. are two fuzzy sets in the universal set of U with membership
Every input has a weight assigned to it and these weights functions of µA and µB, then:
vectors change by learning rules. Definition 1- Union: The membership function µ(A∪B) of the
union A∪B for all u∈U is defined as followed:
µ(A∪B)=max {µA (u),µB (u)} (4)

Definition 2- Intersection: The membership function µ(A∩B)

of the intersection A∩B for all u∈U is defined as followed:
µ(A∩B)=min {µA (u),µB (u)} (5)

Definition 3- Complement: The membership function µᾹ of

the complement A for all u∈U is defined as followed:

Figure 1. A sample processing element µᾹ (u)=1-µA (u) (6)

There can be a question of how such a large dynamic 4.2. Architecture of Fuzzy Logic Controllers
system processes its information? The answer to this question “Fig. 3” shows the basic configuration of fuzzy logic
is the use of energy functions for the system. Each non-linear controllers which is consisted of four parts.
dynamic system has several equilibrium points. These points
are the minimum points on the energy landscape. If an
arbitrary input pattern is given to the system as its initial state,
60 Fatemeh Zahedi and Zahra Zahedi: A Review of Neuro-Fuzzy Systems Based on Intelligent Control

moves to the margins. Basic rules are usually determined

using fuzzy clustering algorithms. The neural network
approximates the membership functions from the training data.
“Fig. 4” shows a neuro-fuzzy cooperative model.
An example of a neuro-fuzzy structure is as follows [1], [2]:
As it can be seen in “Fig. 5”, this structure contains five
layers. Two linguistic nodes are present for each output, one
for the training data (the desired output), and another for the
actual output of the neuro-fuzzy system.
Figure 3. Basic configuration of fuzzy logic controllers

1- Fuzzification interface which is consisted of the

following steps:
Measuring the value of input variables.
Transfer of a wide range of input variable values to the
related universal set.
Fuzzification, which is in fact the conversation of input
data to proper linguistic value.
2- Knowledge base, including “data base” and “linguistic
(fuzzy) control rule base”.
The data base provides the necessary definitions that are
used to define linguistic control rules and fuzzy data.
The rule base which is specifies the control goals using a
set of linguistic control rules.
3- Decision making logic, which is the kernel of the fuzzy
logic controller. In this part human decision making is
simulated based on fuzzy concepts and inferring fuzzy control
actions by using fuzzy implications and rules of inference. Figure 5. The five layer neuro-fuzzy structure
4- Defuzzification interface which is consisted of the
following steps: The first internal or hidden layer is in charge of the
Conversion of a wide range of output variables to fuzzification of the input variables. Each node in this layer can
universal set. be either singular which identifies a simple membership
Deffuzification of rules and linguistic variables in a way function, or it can be consisted of multi-layered nodes which
that is retrievable and understandable by the rest of calculate the complex membership functions.
process. The second hidden layer defines the preconditions and rules
resulted from the third hidden layer. In this method, the output
is generated using the hybrid-learning algorithm consisted of
5. Neuro-Fuzzy unsupervised learning to retrieve the initial membership
5.1. Neuro-Fuzzy Structure functions and rule base.
5.2. Fuzzy Control of an Unmanned Vehicle

Figure 4. Neuro-fuzzy cooperative model

Neuro-fuzzy is a model in which the neural network uses

the training data to determine the membership functions and
fuzzy rules of the fuzzy logic system. After specifying the Figure 6. The state variables of the fuzzy vehicle
parameters of the fuzzy logic system, the neural network
 Journal of Electrical and Electronic Engineering 2015; 3(2-1): 58-61 61

In this section, an example is presented to show the main

application of this model [1]. The vehicle uses examples to
learn to automatically move through a track with rectangular
turns. X0, X1 and X2 are the linguistic variables that specify the
distance from track boundaries from one side and the current
angle of the steering wheel. The output linguistic variable Y is
the next steering angle. This concept is better explained in
“Fig. 6”.
“Fig. 7”, shows the membership functions of the x0, x1, x2
and y variables, learned after one or two training phases.
“Table, 1” shows the learned fuzzy logic rules. These rules
show the hidden layers of the neural network design. In this
example, we assume that the vehicle moves at a constant
speed and sensors are installed on the vehicle for measuring
the x0, x1 and x2 variables. These variables are given to the Figure 7. The learned membership functions
controller so that it can derive the next steering angle from
them. This example has been simulated under different initial
conditions for the steering angle and the results have been very
close to reality.
Table 1. Learned fuzzy logic rules in this example

Learned fuzzy logic rules

Precondition Consequence Precondition consequence
Rule Rule
x0 x1 x2 Y x0 x1 x2 Y
0 0 1 0 6 15 1 2 1 4
1 1 1 0 6 16 2 2 1 2
2 2 1 0 5 17 0 3 1 2
3 0 2 0 3 18 1 3 1 2
4 1 2 0 2 19 2 3 1 2
5 2 2 0 1 20 0 0 2 11
6 0 3 0 2 21 1 0 2 10
7 1 3 0 2 22 2 0 2 12
8 2 3 0 0 23 0 1 2 10
9 0 0 1 9 24 1 1 2 8
10 1 0 1 9 25 2 1 2 13
11 0 1 1 7 26 0 2 2 7
12 1 1 1 7 27 1 2 2 7
13 2 1 1 7 28 2 2 2 13
14 0 2 1 6 29 0 3 2 13

control and decision system,” IEEE. Trans. Comput, vol. 40, pp.
6. Discussion and Conclusion 1320-1336, December 1991.

In this paper, intelligent control and two of its basic [2] A. Abraham. “Neuro fuzzy systems: state-of-the-art modeling
methods, neural networks and fuzzy logic were generally techniques, connectionist models of neurons,” LNCS, vol.
2084, pp. 269-276, Jun 2001
represented. Cooperative neuro-fuzzy model was reviewed
and the general structure of one of its methods was explained. [3] C. J. Harris, C. G. Moore, and M. Brown. “Intelligent control:
An example was provided for the two main applications of the aspects of fuzzy logic and neural network,” World Scientific
model, fuzzy logic controllers and fuzzy decision making Series in Robotics and Automated Systems. UK, vol. 6, 1993.
systems. This example shows the superiority of the hybrid [4] B. Bavarian. “Introduction to neural networks for intelligent
learning method over the more traditional methods. control,ˮ IEEE Control System Magazine, California, vol. 8, pp.
3-7, April 1988.
[5] C. C. Lee. “Fuzzy logic in control systems: fuzzy logic
References controller- part I,ˮ IEEE. Trans. Syst. Man. Cyber. Syst.
California, vol. 20, pp. 404-418, Mar/Apr 1990.
[1] C. T. Lin, and C.S.G. Lee. “Neural network based fuzzy logic