Date post: | 10-Apr-2018 |
Category: |
Documents |
Upload: | nikhil-mahajan |
View: | 221 times |
Download: | 0 times |
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 1/19
PATTERN RECOGNITION BY
ARTIFICIAL NEURAL NETWORKING (ANN)
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 2/19
Project Assigned
Detection of faulty flow meters through patternrecognition using ANN.
Detection of Oil leaks in pipeline through patternrecognition using ANN
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 3/19
Pattern Recognition
Pattern recognition is "the act of taking in raw dataand taking an action based on the category of the
pattern´
Used to classify data (patterns) based either onprior knowledge or on statistical information
extracted from the patterns.
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 4/19
Applications of Pattern Recognition
REACTOR MODELING THROUGH IN SITU ADAPTIVE LEARNING
PREDICTING PLANT STACK EMISSIONS TO MEET ENVIRONMENTAL
LIMITS
PREDICTING FOULING / COKING IN FIRED HEATERS
PREDICTING OPERATIONAL CREDITS
FORECASTING PRICE CHANGES OF A COMPOSITE BASKET
OF COMMODITIES
CORPORATE DEMOGRAPHIC TREND ANALYSIS
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 5/19
Wavelet transform
The input data is collected by SCADA ( supervisorycontrol and data acquisition)This data is fed to Wavelet Transform
A wavelet is a small wave which oscillates and decaysin the time domain
´ ¹ º ¸
©ª¨
!=! t
x x dt s
t
t x s s sCWT X
] X X ] ] 1
),(),(
Continuous wavelettransform of the signal
x(t) using the analysiswavelet ] (.)
S cale = 1/frequency
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 6/19
Different wavelets are matched with analysiswindow
Analysis windows of different lengths are used fordifferent frequencies:Analysis of high frequencies Use narrower windows forbetter time resolutionAnalysis of low frequencies Use wider windows forbetter frequency resolution
The function used to window the signal is called thewavelet
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 7/19
Few wavelets
Haar Wavelet
Daubechies-4
Mexican Hat
Daubechies-10 Daubechies-40
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 8/19
Applications of wavelet transforms
CompressionDe-noisingFeature ExtractionDiscontinuity DetectionDistribution EstimationData analysis
Biological dataNDE dataFinancial data
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 9/19
Wavelets at work !!
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 10/19
Principal Component Analysis(PCA)
Principal component analysis (PCA) -mathematicalprocedure that transforms a number of possiblycorrelated variables into a smaller number of
uncorrelated variables called principal componentsPCA is mathematically defined as an orthogonal lineartransformation that transforms the data to a newcoordinate system such that the greatest variance byany projection of the data comes to lie on the firstcoordinate (called the first principal component), thesecond greatest variance on the second coordinate, andso on
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 11/19
Computing PCA
Organize the data set- m x n Matrixwhere m ² No. of variables, N ² no of data pointsCalculate the empirical mean ² m x 1 matrix
Calculate the deviations from the meanFind the covariance matrixFind the eigenvectors and eigenvalues of the covariance
matrix ² m x m matrix
Rearrange the eigenvectors and eigenvalues-Sort the columns of the eigenvector matrix and eigenvalues
matrix in order of decreasing eigenvalues
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 12/19
Compute the cumulative energy content for eacheigenvectorSelect a subset of the eigenvectors as basis vectorsConvert the source data to z-scoresProject the z-scores of the data onto the new basis
Y = W*ZWhere, Y - PCA matrix
W*- conjugate transpose of Eigen vector matrixZ ² z score matrix
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 13/19
Introduction To ANN
Artificial neural network (ANN) is a mathematicalmodel or computational model simulating the structureand/or functional aspects of biological neural networks
ANN is an adaptive system -changes its structure basedon external or internal information during learning
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 14/19
Learning
Neural nets mimic human learning processes .Nets are trained iteratively on input data along withthe corresponding target outcomes.After a sufficient number of training iterations, netslearn to recognize patterns creating internal modelsof the processes governing the data.
Two different modes of adaptive learning- Supervised Learning- Unsupervised Learning
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 15/19
Supervised Learning
S upervised learning is a learning technique fordeducing a function from training data.The training data consist of pairs of input objects
(typically vectors), and desired outputsDuring the training process, the differences between theactual output from the net and the desired targetoutcomes) are propagated backwards through the netand are used to update the connecting weights.Repeated iterations of this operation result in a
converged set of weights and a net that has beentrained to identify and learn patterns
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 16/19
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 17/19
Unsupervised Learning
Unsupervised learning involves extraction ofcharacteristic features from a large number ofcases and the subsequent organization of thesecases into groups sharing similar attributesHence in this case , input data points and the
weight functions are provided .This type of learning
leads to cluster formationDifferent techniques used for unsupervised learning
are Radial Basis Function (RBF) and Artificialresonance technique (ART)
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 18/19
Clustering at a glance !
8/8/2019 Pattern Recognition By
http://slidepdf.com/reader/full/pattern-recognition-by 19/19
Clustering