Chapter 1

Fuzzy Sets
1.1 Denitions

Denition 1.1.1 (Fuzzy Set). A Fuzzy set is a set whose elements have degrees of membership. Fuzzy sets are an extension of the classical notion of set (known as a Crisp Set). More mathematically, a fuzzy set is a pair (A, A ) where A is a set and A : A [0, 1]. For all x A, A (x) is called the grade of membership of x. If A (x) = 1, we say that x is Fully Included in (A, A ), and if A (x) = 0, we say that x is Not Included in (A, A ). If there exists some x A such that A (x) = 1, we say that (A, A ) is Normal. Otherwise, we say that (A, A ) is Subnormal. For the sake of convenience, usually a fuzzy set is denoted as: A = A (x1 )/x1 + + A (xn )/xn that belongs to a nite universe of discourse: A {x1 , x2 , . . . , xn } = X where A (xi )/xi (a singleton) is a pair grade of membership element.

Denitions on Fuzzy Sets

We have the following denitions for two fuzzy sets (A, A ) and (B, B ), where A, B X (as crisp sets): Equality: A = B i A (x) = B (x) for all x X Inclusion: A B i A (x) B (x) for all x X
n

Cardinality: |A| =
i=1

A (xi )

Empty Set: A is empty i A (x) = 0 for all x X . -Cut: Given [0, 1], the -cut of A is dened by A = {x X | A (x) }

Operations on Fuzzy Sets

Let (A, A ) and (B, B ) be a fuzzy sets. Complementation: (A, A ), where A = 1 A Height: h(A) = max A (x)
x X

Support: supp(A) = {x X | A (x) > 0} Core: C (A) = {x X | A (x) = 1} Intersection: C = A B , where C = min{A (x), B (x)}
xX

Union: C = A B , where C = max{A (x), B (x)}

x X

Bounded Sum: C = A + B , where C (x) = min{1, A (x) + B (x)}

xX

Bounded Difference: C = A B , where C (x) = max{0, A (x) B (x)}

x X

Exponentiation: C = A where C = (A ) for > 0 Level Set: C = A where C = A for [0, 1] Concentration: C = A where > 1 Dilation: C = A where < 1 Note that if A is crisp, then A = A for all . We will dene the Cartesian product A B to be the same as A B .

Membership Functions
We will usually consider one of the following membership functions: Monotonic Increasing: inc(x; a, b) = max min xa ,1 ,0 ba cx cb ,0 ,0

Monotonic Decreasing: dec(x; b, c) = max min 1, Triangular: tri(x; a, b, c) = max min xa cx , ba cb

Trapezoidal: trap(x; a, b, c, d) = max min 1 2 xc

2

xa dx , ,1 ,0 ba dc

Gaussian: gauss(x; c, ) = exp

II

1.2

Questions

How do you distinguish a fuzzy set from a crisp set? A Fuzzy set allows for elastic membership of its members. Also, the transition from membership to non-membership is gradual, rather than abrupt as for crisp sets. Do the Laws of Contradiction and Excluded Middle hold? Given a Fuzzy set (A, A ), we have The Law of Contradiction: A Ac = The Excluded Middle: A Ac = X However, if A is a non-crisp set, then neither law will hold. Indeed, note that for a non-crisp set, there exists some x A such that A (x) (0, 1), i.e. A (x) = 0, 1. Thus, we have AAc (x) = max{A (x), 1 A (x)} = 0 AAc (x) = min{A (x), 1 A (x)} = 1 Hence, neither law holds for a non-crisp set.

Table 1.1: Operations on a Fuzzy Set Describe the operations on a fuzzy set

III

Chapter 2

Knowledge Representation
2.1
2.1.1

Theory
Fuzzy Systems

A Fuzzy System can be contrasted with a conventional (crisp) system in three main ways: A Linguistic Variable is dened as a variable whose values are sentences in a natural or articial language. Thus, if tall, not tall, very tall, very very tall, etc. are values of height, then height is a linguistic variable. Fuzzy Conditional Statements are expressions of the form IF A THEN B, where A and B have fuzzy meaning, e.g. IF x is small THEN y is large, where small and large are viewed as labels of fuzzy sets. A Fuzzy Algorithm is an ordered sequence of instructions which may contain fuzzy assignment and conditional statements, e.g., x = very small. IF x is small THEN y is large. The execution of such instructions is governed by the compositional rule of inference and the rule of the preponderant alternative.

2.1.2

Knowledge Bases

Once we express our knowledge of a specialism through linguistic variables and rules of thumb that involve imprecise antecedents and consequents, then we have a basis of a knowledge-base. In traditional expert systems facts are stated crisply and rules follow classical propositional logic.

Fuzzication
The membership functions are dened on the input variables and are applied to their actual values to determine the degree of truth for each rule.

Inference
The truth-value for the premise of each rule is computed and the conclusion applied to each part of the rule, resulting in one fuzzy subset assigned to each output variable for each rule. Min and Product are two inference methods. IV

1. In Min inferencing the output membership function is clipped o at a height corresponding to the computed degree of truth of a rules premise. This corresponds to the traditional interpretation of the fuzzy logics AND operation. 2. In Product inferencing the output membership function is scaled by the premises computed degree of truth.

Composition
All the fuzzy subsets assigned to each output variable are combined together to form a single fuzzy subset for each output variable. Max and Sum are two composition rules: 1. In MAX composition, the combined fuzzy subset is constructed by taking the pointwise maximum over all the fuzzy subsets assigned to the output variable by the inference rule. 2. The SUM composition, the combined output fuzzy subset is constructed by taking the pointwise sum over all the fuzzy subsets assigned to output variable by their inference rule. (Note that this can result in truth values greater than 1).

Defuzzication
The fuzzy value produced at the composition stage is converted to a crisp value. Two popular defuzzication techniques are the Centroid and Maximum Method.

Centroid
We can compute the Centre of Gravity (COG) or Centre of Area (COA) of the output of the rules. The COG involves the computation of the weighted sum of the Speed and the corresponding membership function of the output fuzzy set and the weighted sum of the membership function. =
Y

1 x1 ,...,xn (y ) dy
Y

yx1 ,...,xn (y ) dy

The centre of gravity approach attempts to take the rules into consideration according to their degree of applicability. Note that if the graph of is symmetric, then the COG gives = 1 2.

Maximum
Here again the weighted sum and weighted membership are worked out, except that the membership function is given another alpha level cut corresponding to the maximum value of the output fuzzy set. The crisp value for MOM method is given as: = where Y = {y X : (y ) = max } i.e. the MOM is the mean of the the values at which assumes its maximum. V 1 |Y | y
y Y

Recap
Fuzzication: Dene the linguistic variables, their term sets and membership functions based on the information given. Dene the rules of the system. Inference: Dene the membership functions of the rules, recalling the denitions of Fuzzy Sets. Determine which rules are ring for the given input. Composition: If the output membership functions are not specied, combine the membership functions for the ring rules either by taking the overall maximum or the scaled sum (scaled by the sum of the ring degrees). Defuzzication: Use the COG or MOM methods to determine the defuzzied value of the system.

Note
There is no strict set of rules that determine which method above to use. A few guidelines are: For TSK, the product for inference and sum for composition are usually used. There is no de-fuzzication for TSK: the composed value is the nal output value. For inference, the ring rates are on the interval [0, 1], so if our output membership functions are not dened on this interval, it does not make sense to use the min. The product works better in this case. For composition, if our membership functions are dened over a series of intervals, e.g.: x0 0 x 0 x 10 A (x) = 10 1 10 x it can be quite dicult to take the graded sum a series of these. Thus, it is easier to consider the maximum instead. To use the approximate COG or the MOM methods, partition the interval over which your output function is dened and then apply the formulae. Alternatively, the integral form for the COG can be used.

VI

Chapter 3

Fuzzy Control
3.1
3.1.1

Theory
Fuzzy Control

The term control is generally dened as a mechanism used to guide or regulate the operation of a machine, apparatus or constellations of machines and apparatus. Often the notion of control is inextricably linked with feedback: a process of returning to the input of a device a fraction of the output signal. Feedback can be negative, whereby feedback opposes and therefore reduces the input, and feedback can be positive whereby feedback reinforces the input signal. Feedback control is thus a mechanism for guiding or regulating the operation of a system or subsystems by returning to the input of the (sub)system a fraction of the output. The machinery or apparatus etc., to be guided or regulated is denoted by S , the input by W and the output by y , and the feedback controller by C . The input to the controller is the so-called error signal e and the purpose of the controller is to guarantee a desired response of the output y . One can intuitively argue that the control signal, u, in part, is Proportional to the error; Proportional to both the magnitude of the error and the duration of the error Proportional to the relative changes in the error values over time

3.1.2

Conventional Control

In the case of classical operations of process control one has to solve the non-linear function u. Furthermore, it is very important that one also nds the proportionality constants KI , KD , and KP . In the case of fuzzy controller, the non-linear function is represented by a fuzzy mapping, typically acquired from human beings. The above intuition can be expressed more formally as an algebraic equation involving three proportionality constants:
t

u(t) = KP e(t) + KI
0

e( ) d + KD

de dt

VII

Value Proportional (KP ) Integral (KI ) Derivative (KD )

determines reaction to the current error sum of recent errors rate at which the error has been changing

The weighted sum of these three actions is used to adjust the process via a control element such as the position of a control valve or the power supply of a heating element. Conventional control theory uses a mathematical model of a process to be controlled and specications of the desired closed-loop behaviour to design a controller. This approach may fall short if the model of the process: is dicult to obtain, or is (partly) unknown, or is highly non-linear.

3.1.3

Fuzzy Controller

Logical rules with vague predicates can be used to derive inference from vague formulated data. The idea of linguistic control algorithms was a brilliant generalisation of the human experience to use linguistic rules with vague predicates in order to formulate control actions. A Fuzzy Controller is a device that is intended to modelize some vaguely known or vaguely described process. A knowledge-based system for closed-loop control is a control system which enhances the performance, reliability, and robustness of control by incorporating knowledge which cannot be accommodated in the analytic model upon which the design of a control algorithm is based, and that is usually taken care of manual modes of operation, or by other safety and ancillary logic mechanisms. There are two types of fuzzy controllers: Controller Type Mamdani (linguistic) Controller Takagi-Sugeno-Kang Controller Typical Operation Direct closed-loop controller Supervisory controller

The controller can be used with the process in two modes: Feedback mode when the fuzzy controller will act as a control device, and Feed Forward mode where the controller can be used as a prediction device. All inputs to, and outputs from, the controller are in the form of linguistic variables. In many ways, a fuzzy controller maps the input variables into a set of output linguistic variables. Usually, a plant, process, vehicle, or any other object to be controlled is called a system S . The feedback controller is expected to guarantee a desired response, or output y . Regulation is a process described in the control theory literature as a process for keeping the output y close to the setpoint (reference input) w, despite the presence of disturbances, uctuations of the system parameters, and noise measurements. (Error e = w y ). A controller is implemented using the control algorithm. VIII

3.2

Questions

List the key dierences between a Mamdani and a TSK system. A zero-order TSK model van be viewed as a special case of a Mamdani system in which each rule is specied by a fuzzy singleton or some pre-fuzzied consequent. In the TSK model, each rule has a crisp output and the overall-output is obtained by a weighted average. This avoids the time consuming process of defuzzication that is required in a Mamdani model. The weighted average operator is replaced by a weighted sum to reduce the computation further. What is the dierence between an ordinary, piecewise linear t and the TSK method? The piecewise linear t will almost continuously t the whole function and allows us to take a global view of the function. The TSK takes into account the changes in the x and y values, and is a smooth piecewise t. How would you resolve conict in a knowledge-based system? Three ways: Specicity Principle: This states that if a number of rules are applicable to a given situation, then the rule with the greatest number of condition premises should be selected to re. Refractoriness Principle: This states that if a rules has been applied on a previous cycle, then it should not be applied again to the same set of facts in data memories. Recency Principle: This states that if more than one rule applies to a given situation, then choose the rule that applies to the most recently entered items in the data memory. How do you think TSK systems learn from observations? Typically, rules contain membership functions for both antecedents and consequents: Mamdani - if e(k ) 0 and e(k ) 0, then u(k ) 0 TSK - if e(k ) 0 and e(k ) 0, then u(k ) = e(k ) + e(k ) + where , , are obtained from empirical observations by relating the behaviour of the errors and change in errors over a xed range of changes in control. Does a 0-order TSK system show where the vagueness of ideas is coming through? What is the source of the error? A TSK model describes a fuzzy implication as follows: if (x1 is A (x1 ), x2 is A (x2 ), . . . , xn is A (xn ), ) then y = g (x1 , x2 , . . . , xn )

A zero-order TSK model describes a fuzzy implication as follows: if (x1 is A (x1 ), x2 is A (x2 ), . . . , xn is A (xn ), ) then y=k

The vagueness comes from the usage of vague predicates, e.g. e is approximately zero, u is positive, e is negative, etc.

IX