Fuzzy Systems

Fuzzy Sets and Fuzzy Logic

FS – Fuzzy Sets and Fuzzy Logic 1 / 119

What are we going to talk about?!

Research on fuzzy systems wants to establish

• theoretical and methodological bases for computational
intelligence,
• tools and techniques for design of intelligent systems.

Fuzzy systems focus on applications

• where some aspects of imprecision plays an important role.

Fuzzy set theory and fuzzy logic

• with a solid mathematical foundation.

FS – Fuzzy Sets and Fuzzy Logic 2 / 119

Motivation

3 / 119
Motivation
Every day humans use imprecise linguistic terms
e.g. big, fast, about 12 o’clock, old, etc.
All complex human actions are decisions based on such concepts:
• driving and parking a car,
• financial/business decisions,
• law and justice,
• giving a lecture,
• listening to the professor/tutor.
So, these terms and the way they are processed play a crucial role.
Computers need a mathematical model to express and process such
complex semantics.
Concepts in classical mathematics are inadequate for such models.
FS – Fuzzy Sets and Fuzzy Logic 4 / 119
 Lotfi Asker Zadeh (1965)

Classes of objects in the real world do not have

precisely defined criteria of membership.
Such imprecisely defined “classes” play an
important role in human thinking,
Particularly in domains of pattern recognition,
communication of information, and abstraction.
Zadeh in 2004 (born 1921)

FS – Fuzzy Sets and Fuzzy Logic 5 / 119

Imprecision

Any notion is said to be imprecise when its meaning is not fixed by

sharp boundaries.

Can be applied fully/to certain degree/not at all. Gradualness

(“membership gradience”) also called fuzziness. A Proposition
is imprecise if it contains gradual predicates.
Such propositions may be neither true nor false, but in-between.
They are true to a certain degree, i.e. partial truth.
Forms of such degrees can be found in natural language, e.g. very,
rather, almost not, etc.

FS – Fuzzy Sets and Fuzzy Logic 6 / 119

Example – The Sorites Paradox

If a sand dune is small, adding one grain of sand to it leaves it small. A

sand dune with a single grain is small.
Hence all sand dunes are small.

Paradox comes from all-or-nothing treatment of small.

Degree of truth of “heap of sand is small” decreases by adding one
grain after another.
Certain number of words refer to continuous numerical scales.

FS – Fuzzy Sets and Fuzzy Logic 7 / 119

Example – The Sorites Paradox

How many grains of sand has a sand dune at least?

Statement A(n): “n grains of sand are a sand dune.”

Let dn = T (A(n)) denote “degree of acceptance” for A(n).
Then
0 = d0 ≤ d1 ≤ . . . ≤ dn ≤ . . . ≤ 1

FS – Fuzzy Sets and Fuzzy Logic 8 / 119

Imprecision

Consider the notion bald:

A man without hair on his head is bald, a
hairy man is not bald.
Usually, bald is only partly applicable.
Where to set baldness/non baldness threshold?

Fuzzy set theory does not assume any threshold!

FS – Fuzzy Sets and Fuzzy Logic 9 / 119

Uncertainty

Uncertainty describes the probability of a well-defined proposition

Rolling a die will either lead to exactly 6 or not, but not something
around 6

Uncertainty is different from Imprecision

Uncertainty comes e.g. from randomness

There are lots of non-standard calculi for handling uncertainty e.g.:

• possibility theory.

FS – Fuzzy Sets and Fuzzy Logic 1010/ /119

119
Distinction between Imprecision and Uncertainty

Imprecision:
e.g. “Today the weather is fine.”
Imprecisely defined concepts
neglect of details
computing with words

Uncertainty:
e.g. “How will the exchange rate of the dollar be tomorrow?”
probability, possibility

FS – Fuzzy Sets and Fuzzy Logic 11 / 119

Examples of Imprecision and Uncertainty
Uncertainty differs from imprecision. It can result from it.

“This car is rather old.” (imprecision)

Lack of ability to measure or to evaluate numerical features.

“This car was probably made in Germany.” (uncertainty)

Uncertainty about well-defined proposition made in Germany,
perhaps based on statistics (random experiment).

“The car I chose randomly is perhaps very big.” (uncertainty and

imprecision)
Lack of precise definition of notion big.
Modifier very indicates rough degree of “bigness”.

FS – Fuzzy Sets and Fuzzy Logic 12 / 119

Lotfi A. Zadeh’s Principle of Incompatibility

“Stated informally, the essence of this principle is that as the

complexity of a system increases, our ability to make precise
and yet significant statements about its behavior diminishes
until a threshold is reached beyond which precision and
significance (or relevance) become almost mutually exclusive
characteristics.”

Fuzzy sets/fuzzy logic are used as mechanism for abstraction of

unnecessary or too complex details.

FS – Fuzzy Sets and Fuzzy Logic 13 / 119

Applications of Fuzzy Systems
Control Engineering
Approximate Reasoning
Data Analysis
Image Analysis

Advantages:
Use of imprecise or uncertain information
Use of expert knowledge
Robust nonlinear control
Time to market
Marketing aspects

FS – Fuzzy Sets and Fuzzy Logic 14 / 119

Fuzzy Sets

15 / 119
Membership Functions
Lotfi A. Zadeh (1965)
“A fuzzy set is a class with a continuum of membership grades.”

An imprecisely defined set M can often be characterized by a

membership function µM.
µM associates real number in [0, 1] with each element x ∈ X.
Value of µM at x represents grade of membership of x in M.
A Fuzzy set is defined as mapping

µ : X → [0, 1].

Fuzzy sets µM generalize the notion of a characteristic function

χ M : X → {0, 1}.

FS – Fuzzy Sets and Fuzzy Logic 16 / 119

Membership Functions
µM(u) = 1 reflects full membership in M.
µM(u) = 0 expresses absolute non-membership in M.
Sets can be viewed as special case of fuzzy sets where only full
membership and absolute non-membership are allowed.
Such sets are called crisp sets or Boolean sets.
Membership degrees 0 < µM < 1 represent partial membership.

µM

0 Age
0 20 40
Representing young in “a young person”
FS – Fuzzy Sets and Fuzzy Logic 17 / 119
Membership Functions

A Membership function attached to a given linguistic description (such as

young) depends on context:
A young retired person is certainly older than young student.
Even idea of young student depends on the user.

Membership degrees are fixed only by convention:

Unit interval as range of membership grades is arbitrary.

FS – Fuzzy Sets and Fuzzy Logic 18 / 119

Membership Functions
Consider again representation for predicate young
There is no precise threshold between
prototypes of young and
prototypes of not young.
Fuzzy sets offer natural interface between linguistic and numerical
representations.

µM

0 Age
0 20 40
Representing young in “a young person”
FS – Fuzzy Sets and Fuzzy Logic 19 / 119
Example – Body Height of 4 Year Old Boys

1 µtall

0.65

h [m]
0
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Membership function for µtall

1.5 m is for sure tall, 0.7 m is for sure small, but in-between?!
Imprecise predicate tall modeled as sigmoid function,
e.g. height of 1.1 m has membership degree of 0.65.
So, height of 1.1 m satisfies predicate tall with 0.65.
FS – Fuzzy Sets and Fuzzy Logic 20 / 119
Examples for Fuzzy Numbers

1 µ b 1 µ

IR IR
0 0
0 1 2 3 4 0 1 2 3 4
exactly two around two

Exact numerical value has membership degree of 1.

Left: monotonically increasing, right: monotonically decreasing,
i.e. unimodal function.
Terms like around modeled using triangular or Gaussian function.

FS – Fuzzy Sets and Fuzzy Logic 22 / 119

Representation of Fuzzy Sets

22 / 119
Extension to a fuzzy set
ling. description model
all numbers smaller 1 ) characteristic
than 10 objective function of a
[ set
10
all numbers almost 1 membership
equal to 10 subjective function of a
“fuzzy set”
10

Definition
A fuzzy set µ of X ≠∅is a function from the reference set X to the unit
interval, i.e. µ : X → [0, 1]. F( X ) represents the set of all fuzzy
sets of X, i.e. F( X ) d=ef {µ | µ : X → [0, 1]}.
FS – Fuzzy Sets and Fuzzy Logic 29 / 119
Vertical Representation
So far, fuzzy sets were described by
their characteristic/membership function and
assigning degree of membership µ(x) to each element x ∈ X.
That is the vertical representation of the corresponding fuzzy set,
e.g. linguistic expression like “about m”

or “approximately between b and c”

FS – Fuzzy Sets and Fuzzy Logic 30 / 119

Horizontal Representation

Another representation is very often applied as follows:

For all membership degrees α belonging to chosen subset of [0, 1], human
expert lists elements of X that fulfill vague concept of fuzzy set with
degree ≥ α.
That is the horizontal representation of fuzzy sets by their α-cuts.

Definition
Let µ ∈ F( X ) and α ∈ [0, 1]. Then the sets

[µ]α = {x ∈ X | µ(x) ≥ α} , [µ]α = {x ∈ X | µ(x) > α}

are called the α -cut and strict α -cut of µ.

FS – Fuzzy Sets and Fuzzy Logic 31 / 119

A Simple Example

Let A ⊆ X, χ A : X → [0, 1]

Then [χ A ] α = A for 0 < α ≤ 1.

χ A is called indicator function or characteristic function of A.

FS – Fuzzy Sets and Fuzzy Logic 32 / 119

An Example

1
µ
α

0
a m b IR
[µα]
Let µ be triangular function on IR as shown above.
α-cut of µ can be constructed by
1. drawing horizontal line parallel to x-axis through point (0, α),
2. projecting this section onto x-axis.

FS – Fuzzy Sets and Fuzzy Logic 33 / 119

Properties of α-cuts I

Any fuzzy set can be described by specifying its α-cuts.

That is the α-cuts are important for application of fuzzy sets.

Theorem
Let µ ∈ F( X ), α ∈ [0, 1] and β ∈ [0, 1].

FS – Fuzzy Sets and Fuzzy Logic 34 / 119

Properties of α-cuts II
Theorem (Representation Theorem)
Let µ ∈ F( X ). Then

So, fuzzy set can be obtained as upper envelope of its α-cuts.

Simply draw α-cuts parallel to horizontal axis in height of α.

FS – Fuzzy Sets and Fuzzy Logic 35 / 119

“Approximately 5 or greater than or equal to 7”
An Exemplary Horizontal View
Suppose that X = [0, 15].
An expert chooses L = {0, 0.25, 0.5, 0.75, 1} and α-cuts:
• A0 = [0, 15],
• A0.25 = [3, 15],
• A0.5 = [4, 6] ∪ [7, 15],
• A0.75 = [4.5, 5.5] ∪ [7, 15],
• A1 = {5} ∪ [7, 15].
1.00
0.75
0.50
0.25
0
0 5 10 15
The family (Aα)α∈L of sets induces upper shown fuzzy set.
FS – Fuzzy Sets and Fuzzy Logic 38 / 119
“Approximately 5 or greater than or equal to 7”
An Exemplary Vertical View

µ A is obtained as upper envelope of the family A of sets.

The difference between horizontal and vertical view is obvious:
1.00
0.75
0.50
0.25
0
0 5 10 15
The horizontal representation is easier to process in computers.
Also, restricting the domain of x-axis to a discrete set is usually done.

FS – Fuzzy Sets and Fuzzy Logic 39 / 119

Support and Core of a Fuzzy Set

Definition
The support S(µ) of a fuzzy set µ ∈ F( X ) is the crisp set that contains
all elements of X that have nonzero membership. Formally

S(µ) = [µ]0 = {x ∈ X | µ(x) > 0} .

Definition
The core C(µ) of a fuzzy set µ ∈ F( X ) is the crisp set that contains all
elements of X that have membership of one. Formally,

C(µ) = [µ]1 = {x ∈ X | µ(x) = 1} .

FS – Fuzzy Sets and Fuzzy Logic 41 / 119

Height of a Fuzzy Set

Definition
The height h(µ) of a fuzzy set µ ∈ F( X ) is the largest membership
grade obtained by any element in that set. Formally,

h(µ) = sup {µ(x)} .

x∈X

h(µ) may also be viewed as supremum of α for which [µ]α ≠∅.

Definition
A fuzzy set µ is called normal, iff h(µ) = 1. It
is called subnormal, iff h(µ) < 1.

FS – Fuzzy Sets and Fuzzy Logic 42 / 119

Convex Fuzzy Sets

Definition
Let X be a vector space. A fuzzy set µ ∈ F( X ) is called fuzzy convex
if its α-cuts are convex for all α ∈ (0, 1].

FS – Fuzzy Sets and Fuzzy Logic 43 / 119

Fuzzy Numbers

Definition
µ is a fuzzy number if and only if µ is normal and [µ]α is bounded,
closed, and convex Ɐα ∈ (0, 1].

Example:
The term approximately x0 is often described by a parametrized class of
membership functions, e.g.

µ1(x) = max{0, 1 − c1|x − x0|}, c1 > 0,

µ2(x) = exp(−c2ǁx − x0ǁp), c2 > 0, p ≥ 1.

FS – Fuzzy Sets and Fuzzy Logic 44 / 119

 Convex Fuzzy Sets
1 1

α α

0 0
IR IR

Theorem
A fuzzy set µ ∈ F(IR) is convex if and only if

µ(λx1 + (1 − λ)x2) ≥ min{µ(x1), µ(x2)}

for all x1, x2 ∈ IR and all λ ∈ [0, 1].

FS – Fuzzy Sets and Fuzzy Logic 45 / 119

 Fuzzy Numbers – Example

1.0

0.5

0
0 1 2 3 4

Upper semi-continuous functions are often convenient in applications.

In many applications (e.g. fuzzy control) the class of the functions and
their exact parameters have a limited influence on the results.
Only local monotonicity of the functions is really necessary.
In other applications (e.g. medical diagnosis) more precise membership
degrees are needed.
FS – Fuzzy Sets and Fuzzy Logic 46 / 119
Set Operators. . .

. . . are defined by using traditional logics operator

Let X be universe of discourse (universal set):

A ∩ B = {x ∈ X | x ∈ A ∧ x ∈ B}
A ∪ B = {x ∈ X | x ∈ A 𝖵 x ∈ B}
Ac = {x ∈ X | x ∈
/ A} = {x ∈ X | ¬( x ∈ A)}

One idea to define fuzzy set operators: use fuzzy logics.

FS – Fuzzy Sets and Fuzzy Logic 47 / 119

Fuzzy Logic

39 / 119
Propositions and Truth Values
Classical propositional logic deals with the formal handling of statements
(propositions) to which one of the two truth values 1 (for true) or 0 (for
false) can be assigned.

The truth value which is assigned to a proposition ϕ is denoted by [[ϕ]].

The most important logical connectives are the logical AND ∧

(conjunction), the logic OR ∨ (disjunction), the negation NOT ¬ and the
IMPLICATION →.

FS – Fuzzy Sets and Fuzzy Logic 60 / 119

Conjunction, disjunction, implication and negation

Four is an even number AND 2 + 5 = 9.

Four is an even number IMPLICATION 2 +5 = 9.
Four is an even number OR 2 + 5 = 9.
NOT 2 + 5 = 9.
[[ϕ1∧ϕ2]] = 0, [[ϕ1 → ϕ2]] = 0, [[ϕ1 ∨ ϕ2]] = 1 and [[¬ϕ2]] = 1
FS – Fuzzy Sets and Fuzzy Logic 61 / 119
The statement of being at a party at about 5. The greater
the difference between the arrival and 5 o’clock the “less
true” the statement is. A sharp definition of an interval of
time corresponding to “about 5” is impossible.

Humans are able to formulate such “fuzzy” statements,

understand them, draw conclusions from them and work
with them.
A formalization of this simple issue in a logical
calculus, where statements can be either true or false
only, is not adequate.

FS – Fuzzy Sets and Fuzzy Logic 61 / 119

Conjunction, disjunction, implication and negation

Combined propositions like “160 km/h is a high velocity

AND the stopping distance is about 110 m”

Extension of the truth tables for logical connectives

Binary:

Fuzzy:
A minimum requirement we demand of these functions is
that, limited to the values 0 and 1, they should provide
the same values as the corresponding truth function
associated with the connectives of classical logic.
FS – Fuzzy Sets and Fuzzy Logic 61 / 119
Conjunction, disjunction, implication and negation

The most frequently used truth functions

w∧(α,β) = min{α,β}
w∨(α,β) = max{α,β}
w¬(α) = 1 - α

The implication
Łukasiewicz implication

Gödel implication

FS – Fuzzy Sets and Fuzzy Logic 61 / 119

Standard Fuzzy Set Operators
Definition

We define the following algebraicoperators on F( X ):

intersection (“AND”),
union (“OR”),
complement (“NOT”).

µ is subset of µ′ if and only if µ ≤ µ′.

FS – Fuzzy Sets and Fuzzy Logic 65 / 119

Standard Fuzzy Set Operators – Example

fuzzy complement two fuzzy sets

fuzzy intersection fuzzy union

FS – Fuzzy Sets and Fuzzy Logic 61 / 119

Fuzzy Set Intersection and Union

47 / 119
Fuzzy Set Intersection and Union

Let A, B be fuzzy subsets of X, i.e. A, B ∈ F( X ).

Their intersection and union can be defined pointwise using:

(A ∩ B)(x) = 𝖳(A(x), B(x )) where 𝖳 : [0, 1]2 → [0, 1]

(A ∪ B)(x) = ⊥(A(x), B(x)) where ⊥ : [0, 1]2 → [0, 1].

48 / 119
Triangular Norms and Conorms I

𝖳 is a triangular norm (t-norm) ⇐ ⇒ 𝖳 satisfies conditions T1-T4

⊥ is a triangular conorm (t-conorm) ⇐ ⇒ ⊥ satisfies C1-C4

for all x, y ∈ [0, 1], the following laws hold

Identity Law
T1: 𝖳( x , 1) = x (A ∩ X = A)
C1: ⊥(x, 0) = x (A ∪ ∅ = A).

Commutativity
T2: 𝖳( x , y) = 𝖳( y , x) (A ∩ B = B ∩ A),
C2: ⊥(x, y) = ⊥(y, x) (A ∪ B = B ∪ A).

49 / 119
Triangular Norms and Conorms II

for all x, y, z ∈ [0, 1], the following laws hold

Associativity
T3: 𝖳( x , 𝖳( y , z)) = 𝖳(𝖳( x , y), z) (A ∩ (B ∩ C)) = ((A ∩ B) ∩ C),
C3: ⊥(x, ⊥(y, z)) = ⊥(⊥(x, y), z) (A ∪ (B ∪ C)) = ((A ∪ B) ∪ C).

Monotonicity
y ≤ z implies
T4: 𝖳( x , y) ≤ 𝖳( x , z)
C4: ⊥(x, y) ≤ ⊥(x, z).

FS – Fuzzy Sets and Fuzzy Logic 76 / 119

Triangular Norms and Conorms III

𝖳 is a triangular norm (t-norm) ⇐ ⇒ 𝖳 satisfies conditions T1-T4

⊥ is a triangular conorm (t-conorm) ⇐ ⇒ ⊥ satisfies C1-C4

for any t-norm 𝖳 :𝖳( x , y) ≤ min(x, y),

for any t-conorm ⊥ :⊥(x, y) ≥ max(x, y).

note: x = 1 ⇒ T (0, 1) = 0 and

x ≤ 1 ⇒ T (x, 0) ≤ T (1, 0) = T (0, 1) = 0

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

Examples of t-norms

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

Examples of t-norms
The limits are given by the drastic product, which is
the smallest t-norm.
Minimum is the only idempotent t-norm which means
that only the minimum satisfies the property t(α,α) =
α for all α ∈ [0, 1].
guarantee that the truth values of the proposition ϕ
and ϕ ∧ ϕ coincide.

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

Idempotency property is not always desirable
A buyer has to decide between the houses A and B.

He chooses house x ∈ {A, B} for which the proposition

“The price of house x is good AND The location of house
x is good” yields the greater truth value.
FS – Fuzzy Sets and Fuzzy Logic 77 / 119
t-norms and t-conorms are dual concepts
Each t-norm induces a t-conorm s by

De Morgan’s Laws

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

Examples of t-conorms

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

A continuous t-norm t induces the
residuated implication t, by the formula

We can obtain by residuation the Łukasiewicz implication

from the Łukasiewicz t-norm and the Gödel implication
from the minimum.

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

 The quantifiers ∀ (all) and ∃ (exists)
Discourse X and the predicate P (x)

X could be the set {2, 4, 6, 8, 10}

P (x) is the predicate “x is an even number”

If the set X is finite, e.g. X = {x1,...,xn}, the statement

(∀x ∈ X)(P (x)) is equivalent to the statement
P (x1) ∧ ··· ∧ P (xn).

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

 The quantifiers ∀ (all) and ∃ (exists)
If we assign the minimum to the conjunction as truth value
function

extended to an infinite universe of discourse X.

Other t-norms than the minimum are normally not used for
the universal quantifier, since the non-idempotent property
leads easily to the truth value zero in the case of
an infinite universe of discourse.
FS – Fuzzy Sets and Fuzzy Logic 77 / 119
 The quantifiers ∀ (all) and ∃ (exists)
The existential quantifier

If the universe of discourse for the existential quantifier is

finite, the propositions ∃x ∈ X : P (x) and P (x1) ∨ · · · ∨ P
(xn) are equivalent.

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

 The quantifiers ∀ (all) and ∃ (exists)
Example:
The predicate P(x) “x is a high velocity”.
The truth value [[P (x)]]
[[P (150)]] = 0,
[[P (170)]] = 0.5
[[P (190)]] = 1.

The statement ∀x ∈ [170, 200] : P (x) (“All velocities

between 170 km/h and 200 km/h are high velocities”)

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

The quantifiers ∀ (all) and ∃ (exists)

FS – Fuzzy Sets and Fuzzy Logic 77 / 119

The Minimum

𝖳 min and ⊥max can be easily processed numerically and visually,

e.g. linguistic values young and approx. 20 described by µy, µ20.
𝖳min(µy , µ20) is shown below.

1
µy µ20

µy ∩ µ20
0
0 10 20 30 40 50

FS – Fuzzy Sets and Fuzzy Logic 8163/ /119

119
Examples of Fuzzy Intersections

t-norm 𝖳 min t-norm 𝖳prod

t-norm 𝖳Łuka

FS – Fuzzy Sets and Fuzzy Logic 86 / 119

Examples of Fuzzy Unions

t-conorm ⊥max t-conorm ⊥sum

t-conorm ⊥Łuka

FS – Fuzzy Sets and Fuzzy Logic 87 / 119

 Complement
Like the complement for ordinary sets, the complementing of
fuzzy sets is an involution, which means that holds
In classical set theory, we have that the intersection of a set with
its complement yields the whole universe of discourse.

FS – Fuzzy Sets and Fuzzy Logic 87 / 119

 Linguistic Modifiers

specific unary operations other than complement

Normally, a fuzzy set represents a vague concept like “high
velocity”, “young” or “tall”. From such concepts, we can
derive other vague concepts applying linguistic modifiers
like “very” or “more or less”.

FS – Fuzzy Sets and Fuzzy Logic 87 / 119