You are on page 1of 19

Fuzzy system applications

(sections 3.8- 3.12, pp.223-247)

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Fuzzy system applications

Pattern recognition and classification Fuzzy clustering Image and speech processing Fuzzy systems for prediction Fuzzy control Monitoring Diagnosis Optimisation and decision making Group decision making

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Pattern recognition and classification

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Pattern recognition and classification


A2 Large Class 1 Class 2 Medium Small

A1 Small Medium Large

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Pattern recognition and classification


RULE 1: IF Nucleus might exist and cell might be segmented and granular and Nucleus is dark, THEN the cell is unlikely to be an Erythrocyte and the cell is possibly a Lymphocyte. RULE 2: IF Nucleus might exist and cell might be segmented and granular and Nucleus is light THEN the cell is unlikely to be an Erythrocyte and the cell is unlikely to be a Lymphocyte. RULE 3: IF Nucleus might exist and cell is not segmented and granular and Nucleus is light, THEN the cell maybe an Erythrocyte and the cell is unlikely to be a Lymphocyte and the cell is likely to be a Monocyte. RULE 4: IF Nucleus might exist and cell is not segmented and granular and Nucleus is dark, THEN the cell is unlikely to be an Erythrocyte and the cell is unlikely to be a Monocyte and the cell is likely to be a Lymphocyte. RULE 5: IF Nucleus exists and the cell might be segmented and granular and the nucleus is light THEN the cell is likely to be a Monocyte. RULE 6: IF Nucleus exists and the cell might be segmented and granular and the Nucleus is dark THEN the cell is likely to be a Lymphocyte. RULE 7: IF Nucleus exists and the cell is not segmented and granular and the Nucleus is dark THEN the cell is a Lymphocyte. RULE 8: IF Nucleus exists and the cell is not segmented and granular and the Nucleus is light THEN the cell is a Monocyte. RULE 9: IF Nucleus is large and granules are large and Nucleus is medium dark THEN the cell is likely to be a Basophil. RULE 10: IF Nucleus is large and granules are large and Nucleus is light THEN the cell is Basophil. RULE 11: IF Nucleus is large and granules are medium and Nucleus is light THEN the cell is likely to be a Basophil.

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Fuzzy clustering
; fact template ;(account <number> <money> <holder>) (deffacts initiallist (account 1234 45.67 smith) (account 3421 0.0 feldman) (account 3333 0.0 grisman)) (defrule printt (account ?num ?money&:(= ?money 0.0) ?name) => (printout t "clear account " ?num " " ?name crlf ))

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Fuzzy clustering
1. Initialise c fuzzy cluster centers V1, V2,..., Vc arbitrarily and calculate the membership degrees i,k, i=1,2,...,c, k=1,2,...,n such that the general conditions are met. 2. Calculate thennext values for cluster centres: n Vi = ( (i,k)2.xk) / ((i,k)2), for i=1,2,...,c k=1 k=1 3. Update the fuzzy degree of membership: 1 i,k = ,for di,k > 0, i,k c dik [ ] djk j=1 where: di,k = (xk - Vi)2, dj,k = (xk - Vj)2 (Euclidean distance) 4. If the currently calculated values Vi for the cluster centers are not different from the values calculated at the previous step (subject to a small error ), then stop the

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Low Medium High

Image and speech processing


30 (a) IF 0-1000 is Medium AND 1000-2000 is Medium AND 2000-3000 is Low THEN the note is Middle C 50

dB

900 800

New Zealand English General Australian English R.P. English

IF 0-1000 is High AND 1000-2000 is Medium AND 2000-3000 is Low THEN the note is D above Middle C IF 0-1000 is High AND 1000-2000 is Low AND 2000-3000 is Low THEN the note is E above Middle C IF 0-1000 is Medium AND 1000-2000 is Medium AND 2000-3000 is Medium THEN the note is F above Middle C IF 0-1000 is Low AND 100-2000 is Medium AND 2000-3000 is Medium AND 3000-4000 is Low THEN the note is G above Middle C

F1 frequency (Hertz)

600 500 400 u 300 200 0 500 1000

3 3

I u u
3

I i

i i

IF 0-1000 is Low AND 1000-2000 is Medium AND 2000-3000 is Medium AND 3000-4000 is Medium THEN the note is A above Middle C IF 0-1000 is Low AND 1000-2000 is Medium AND 2000-3000 is Low THEN the note is B above Middle C

1500 F2 frequency (Hertz)

2000

2500

IF 0-1000 is High AND 1000-2000 is High AND 2000-3000 is High THEN the note is C above Middle C (b)

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Fuzzy systems for prediction


1 FF1 1 for Ra Rs

0 Low Normal High

FF

0 Low Normal High US short term interest

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Fuzzy systems for prediction


PL PM PS ZR NS NM NL NL NM NS ZR PS PM PL NM NS ZR ZR ZR PS PM NL NM NS ZR PS PM PL

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

() (Motor current)

Fuzzy control
Fuzzification X

Fuzzy inference

B Defuzzification rules

Y Process

IF X is A THEN Y is B

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Fuzzy control

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Fuzzy control
PL PM PS ZR NS NM NL NL NM NS ZR PS PM PL NM NS ZR ZR ZR PS PM NL NM NS ZR PS PM PL

N. Kasabov Foundations of Neural Networks, Fuzzy

() Systems, and Knowledge (Motor current)

Engineering, MIT Press, 1996

Fuzzy control
V2 A2 V1 A1 +V Motor controller Fuzzy Controller 1 Stage 1 Fuzzy Controller 2 Motor

Controller One (Output: Stage 1) Velocity 2

Controller Two (Output: Motor) Stage 1 = ZR Zero Velocity 1

Angle 2

NL NL NL

NM ZR NL NL NL

PM

PL

Angle 1

NL NL NL

NM ZR NL NL NL

PM

PL

NM NS PS PM PL PL PL

NM NS PS PM PL PL PL

NM NL NS PS PM PL NL

NM NS PS PM PL PL

NM NL NS PS PM PL NL

NM NS PS PM PL PL

NM NS PS PM PL

NM NS PS PM PL

NM NS NS PS PS PM

NM NS NS PS PS PM

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Monitoring
Normal Quick Normal Slow Underheating Overheating

10

50

90

120

150

Brakes' response (seconds)

Cooling system (t)

Damaged OK Low Normal High

50

90

120

150

Gauge sensitivity (levels)

Temperature (C)

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Diagnosis
Fuzzy rules: Rule 1:IF(M1 is Always AND M2 is Weak AND M3 is No AND M4 is No) THEN D is D1 (CF is very strong); Rule 2:IF (M1 is No AND M2 is Always AND M3 is Weak AND M4 is No) THEN D is D2 (CF is very strong); Rule 3:IF(M1 is More_or_less_weak AND M2 is No AND M3 is Always AND M4 is No) THEN D is D3 (CF is very strong); Rule 4:IF(M1 is Weak AND M2 is More_or_less_weak AND M3 is Always AND M4 is No) THEN D is D4 (CF is very strong);

Use of typical values instead of fuzzy values: Rule 1: IF (M1=1.0, M2=0.2, M3=0.0, M4=0.0) THEN D1 (CF=0.99); Rule 2: IF (M1=0.0, M2=1.0, M3=0.2, M4=0.0) THEN D2 (CF=0.99); Rule 3: IF (M1=0.3, M2=0.0, M3=1.0, M4=0.0) THEN D3 (CF=0.99);
Rule 4: I F (M1 =0.2, M2=0.3, M3=1.0 , M4=0.0 ) THEN D4 (CF=0.99);

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Diagnosis
Fuzzy label Always Very strong Strong More or less strong Medium More or less weak Weak Very weak No Numerical interval [1.00, 1.00] [0.95, 0.99] [0.80, 0.94] [0.65, 0.79] [0.45, 0.64] [0.30, 0.44] [0.10, 0.29] [0.01, 0.09] [0.00, 0.00] Typical value 1.0 0.99 0.9 0.7 0.5 0.3 0.2 0.05 0.0

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Optimisation and decision making

fig 3.52

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996

Group decision making


A

group of experts A set of options Parameters describing the experts' opinions and preferences

N. Kasabov Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering, MIT Press, 1996