Professional Documents
Culture Documents
12 Deepak Fuzzy Logic SVM PDF
12 Deepak Fuzzy Logic SVM PDF
Logic Control
Outline
General Definition
Applications
Operations
Rules
Fuzzy Logic Toolbox
FIS Editor
Tipping Problem: Fuzzy Approach
Defining Inputs & Outputs
Defining MFs
Defining Fuzzy Rules
2
General Definition
Fuzzy Logic - 1965 Lotfi Zadeh, Berkely
Applications
Expert Systems
Control Units
Bullet train between Tokyo and
Osaka
Video Cameras
Automatic Transmissions
Operations
AB
AB
Controller Structure
Fuzzification
Scales and maps input variables to fuzzy sets
Inference Mechanism
Approximate reasoning
Deduces the control action
Defuzzification
Convert fuzzy output values to control signals
10
11
12
13
Input/Output
14
15
16
17
18
19
Membership Functions
20
Rules
21
22
Rule Base
23
Rule Viewer
24
Surface Viewer
25
26
27
28
29
30
Add Input MF
31
Define Input MF
32
Add output MF
33
Define output MF
34
Add rules
35
36
View rules
37
Rules viewer
38
Surface viewer
39
It is intuitive.
It has widespread acceptance.
It is well suited to human input.
41
Overview
Properties of SVM
Applications
Discussion
Linear Classifiers
w x + b>0
denotes +1
yest
denotes -1
f(x,w,b) = sign(w x + b)
w x + b<0
Linear Classifiers
x
denotes +1
denotes -1
f(x,w,b) = sign(w x + b)
yest
Linear Classifiers
x
denotes +1
f
f(x,w,b) = sign(w x + b)
denotes -1
yest
Linear Classifiers
x
yest
f(x,w,b) = sign(w x + b)
denotes +1
denotes -1
Any of these
would be fine..
..but which is
best?
Linear Classifiers
x
denotes +1
f
f(x,w,b) = sign(w x + b)
denotes -1
Misclassified
to +1 class
yest
Classifier Margin
x
denotes +1
denotes -1
yest
f(x,w,b) = sign(w x + b)
Maximum Margin
denotes +1
denotes -1
Support Vectors
are those
datapoints that
the margin
pushes up
against
margin linear
3. Empirically it works
very very
classifier
iswell.
the
linear classifier
with the, um,
maximum margin.
Linear SVM
This is the
simplest kind of
SVM (Called an
LSVM)
M=Margin Width
X-
What we know:
x+
w.
+ b = +1
w . x- + b = -1
w . (x+-x-) = 2
(x x ) w 2
M =
=
w
w
1 t
Minimize ( w) = w w
2
subject to
yi ( wxi + b) 1
Quadraticoptimizationproblemsareawellknownclassofmathematical
programmingproblems,andmany(ratherintricate)algorithmsexistfor
solvingthem.
ThesolutioninvolvesconstructingadualproblemwhereaLagrange
multiplier iisassociatedwitheveryconstraintintheprimaryproblem:
Find 1N such that
Q() =i - ijyiyjxiTxj is maximized and
(1)
i y i = 0
denotes -1
- No training error
OVERFITTING!
11
Linear SVMs:Overview
Non-linear SVMs
: x (x)
K=
K(x1,xN)
K(x2,xN)
K(xN,xN)
K (x i , x j ) = exp(
xi x j
2
Properties of SVM
Flexibility in choosing a similarity function
Sparseness of solution when dealing with large data sets
- only support vectors are used to specify the separating
hyper plane
Ability to handle large feature spaces
- complexity does not depend on the dimensionality of the
feature space
Over fitting can be controlled by soft margin approach
Nice math property: a simple convex optimization problem
which is guaranteed to converge to a single global solution
Feature Selection
SVM Applications
Weakness of SVM
It is sensitive to noise
- A relatively small number of mislabeled examples can
dramatically decrease the performance
Some Issues
Choiceofkernel
Choiceofkernelparameters
Gaussianorpolynomialkernelisdefault
ifineffective,moreelaboratekernelsareneeded
domainexpertscangiveassistanceinformulating
appropriatesimilaritymeasures
inGaussiankernel
isthedistancebetweenclosestpointswithdifferent
classifications
Optimizationcriterion Hardmarginvs.Softmargin
Alengthyseriesofexperimentsinwhichvarious
parametersaretested
WindPowerForecasting(WPF)
WPFisatechniquewhichprovidestheinformationof
howmuchwindpowercanbeexpectedatagiven
pointoftime.
Duetotheincreasingpenetrationofwindpowerinto
theelectricpowergrid.
Agoodshorttermforecastingwillensuregrid
stabilityandafavorabletradingperformanceonthe
electricitymarkets.
-SVM
StructureofSVM
Data Resolution
Theresolutionofthedatasetis10minutes.
Eachdatarepresentstheaveragewindspeedand
powerwithinonehour.
Thedatavaluesbetweentwoadjacentsamplesare
xj
linearlychanged,thatis:
xi+1 + xi
)
x j (t ) = xi +
.t
dti
0 t dti
xi
xi +1
Whereisthetimeintervalbetweenand.
dti
DataValue
Theaveragevalueofthedatawithincanbe
Ts
calculatedas
1
)
x j (t ) =
Ts
ti +Ts
)
x j (t )dt
ti
where Ts =60minutesisusedintheveryshortterm
Ts
forecasting(lessthan6hours)and=2hoursisused
forshorttermforecasting.
FixedStepPredictionScheme
Predictionhorizonofhsteps
fixedstepforecastingmeansonlythevalueofthe
nextsampleispredictedbyusingthehistorical
hth
data.
(t + h) = f (yt, yt-1,,yt-d)
Wheref isanonlinearfunctiongeneratedbySVM
yt +h ispredictedwiththedatabefore(theredblocks),is
yt
yt +h1 predictedwiththedatabefore(thegreenblocks)
yt 1
Windspeednormalization
Autocorrelationsofthewindspeedsamples
SVMmodelandtheRBFmodel
CONCLUSIONS
TheSVMhasbeensuccessfullyappliedtothe
problemsofpatternclassification,particularlythe
classificationoftwodifferentcategoriesofpatterns.
SVMmodelismoresuitableforveryshorttermand
shorttermWPF
ProvidesapowerfultoolforenhancingtheWPF
accuracy.