Post on 20-Jan-2016
description
1
Network Weather ForecastingMAGGIE(NWF)
Fareena SaqibBIT-4 A(195)
37fareena@niit.edu.pkfareenas@gmail.com
Advisor:
Dr Arshad Ali
Co-Advisor:
Umar kalim
Committee Members:
Aatif Kamal
Kamran hussain
20th July, 2006 Fareena Saqib NWF
Contents
Problem statement Motivation Project Aim Introduction Scope Literature Review Proposed Solution Methodology Project Modules Comparative Analysis Time Line Conclusion Research Accomplishments Future Recommendations
20th July, 2006 Fareena Saqib NWF
Problem Statement
Forecasting the performance of network using technique that better conserves the varying
patterns in the data using historical data collected by different
active monitoring tools
Content
20th July, 2006 Fareena Saqib NWF
Motivation GRID Management System
Allocation of task Parallel processing Storage
GRID System
Task Distributor
Remote
Network1
Remote
Network2
Remote network3
Remote
Network N
Content
20th July, 2006 Fareena Saqib NWF
Project Aim
The aim of project is to develop a module that forecasts the performance of different networks
based on historical data. So that efficiency of the system can be increased.
Content
20th July, 2006 Fareena Saqib NWF
Introduction
Forecasting techniques Holt Winters ARMA/ARIMA EWMA Regression Analysis
Why ARMA/ARIMA? Varying trends in network data
Content
20th July, 2006 Fareena Saqib NWF
ARMA/ARIMA Auto Regressive Integrated Moving Average
Box and Jenkins approach Merger of techniques
o Auto Regression (AR)o Moving Average (MA)
Benefits of AR
Benefits ofMA
ARIMAApproach
Better ResultsContent
20th July, 2006 Fareena Saqib NWF
ARMA/ARIMA Approach followed:
Box and Jenkins approach is followed
1. Identification of the model(Choosing tentative p,d,q)
1. Identification of the model(Choosing tentative p,d,q)
3. Forecasting 3. Forecasting
4. Diagnostic checking (are the estimated residuals white noise?)
4. Diagnostic checking (are the estimated residuals white noise?)
2. Parameter estimation ofthe chosen model
2. Parameter estimation ofthe chosen model
No
(Return to step 1)
Content
20th July, 2006 Fareena Saqib NWF
ARMA/ARIMA Identification
Through Correlogram Autocorrelation Function (ACF) Partial Auto Correlation Function (PACF)
Content
20th July, 2006 Fareena Saqib NWF
ARMA/ARIMA Estimation
Estimation of order Estimation of equation Estimation of coefficients Forecasting of data
Diagnostic Checking To check that model is fit to the data. Obtain residual Obtain ACF and PACF of residual
Content
20th July, 2006 Fareena Saqib NWF
Use of ARIMA Approach
Content
20th July, 2006 Fareena Saqib NWF
Use of ARMA/ARIMA
Sales of dates contains seasonal effect. Month of Ramadan
Sales of products Summer Winter Spring
USA economic forecasts Weather forecasts
Content
20th July, 2006 Fareena Saqib NWF
Network Weather Forecasting
Use of ARIMA in network forecast
Network Weather Forecasting
ComputerScience
Field
EconomicsField
Statistics
EconometricsField
Network
GRID System
Content
20th July, 2006 Fareena Saqib NWF
Scope Study of different forecasting techniques
Pros and cons Selection of Technique Development of methodology Verification of the algorithm Modules:
o Data Processing moduleo Forecasting moduleo Visualization moduleo Testing Moduleo Comparative module o Development of user Interface
Content
20th July, 2006 Fareena Saqib NWF
Research Issues
Research Issues:
Development of algorithm using ARIMA approach Estimation of the coefficients. Diagnostic Checking tests.
Content
20th July, 2006 Fareena Saqib NWF
Literature Review Development of algorithm using ARIMA approach
Basic Econometrics by Damodar N.Gujaratio Basic concepts:o Time Series Analysis:o ARMA and ARIMA approach introduction.
Time Series Analysis Forecasting and Control by George E.P Box, Gwilym M.Jenkins,Gregory C.Reinsel:
o Study of ARMA/ARIMA in detail.o Box and Jenkins Approach
Basics of statisticso To understand and revise basic concepts of statistics involved in the
project.
Research Issues
20th July, 2006 Fareena Saqib NWF
Literature Review Estimation of the coefficients.
Estimation of coefficiento http://www.qmw.ac.uk/~ugte133/courses/tseries/8idntify.pdf
Non-linear approacheso http://www.ece.cmu.edu/~moura/papers/icassp88-ribeiro-ieeexplore.p
df Other approaches
o http://www.cs.cmu.edu/afs/cs/project/cmcl/archive/Remulac-papers/tech-report.pdf
Research Issues
20th July, 2006 Fareena Saqib NWF
Literature Review
Diagnostic Checking tests.
Basic Econometrics by Damodar N.Gujaratio Basic concepts:o Time Series Analysis:o ARMA and ARIMA approach introduction.
Basics of statisticso To understand and revise basic concepts of statistics
involved in the project.
Research Issues
20th July, 2006 Fareena Saqib NWF
Literature Review Algorithms Involved:
Data Processing
•Selection of Parameter
•Trim Operation •Regularization Algorithm •Moving Average for Interpolation
Forecasting
•Stationarity
•Order Estimation
•Coefficient Estimation
•Formulation of equation
Verification
•Calculation of Residuals
•Trend Analysis
•Portmanteau tests
Content
20th July, 2006 Fareena Saqib NWF
Proposed Solution
Content
20th July, 2006 Fareena Saqib NWF
ARIMA Modeling
Postulate General Class of Models
Use Model for Forecasting
Diagnostic Checking
Estimate Parameters in Tentatively
Entertained Model
Identify Model to be Tentatively
Entertained
yesNo
Content
20th July, 2006 Fareena Saqib NWF
Methodology
Read Data Files
Trim Operation
Data Files
RegularizationOperation
Interpolation Operation
Plot of Processed
Data
Content
20th July, 2006 Fareena Saqib NWF
MethodologyProcessed
Data
Read Processed
data in array
Trend Analysis
Order Estimation
Coefficients Estimation
Equation Formulation
Foecasting
Content
20th July, 2006 Fareena Saqib NWF
MethodologyForecasted
Data
Read Forecast
data
Calculate residuals
Trend Analysis
Portmanteau Test
Decision making based on results Content
20th July, 2006 Fareena Saqib NWF
Network Weather Forecasting
EstimationOf order
EstimationOf order
Estimation ofCoefficient
Estimation ofCoefficient ForecastingForecasting
Data TrimData Trim User InterfaceUser Interface
correlogramcorrelogram
Estimatio
n
Access Data
Interpolation Interpolation Regularization
Regularization
CovarianceCovariance
IntegrationIntegration
Identification
Diagnostic Checking
Diagnostic Checking
Content
20th July, 2006 Fareena Saqib NWF
Architecture Diagram
Data files
Docs………………
Docs………………
Docs………………
ForecastingForecasting
User
Data Cleaning
Data Cleaning
EstimationEstimation
GUIGUI
Visualization through graphs
Visualization through graphs
Content
20th July, 2006 Fareena Saqib NWF
Use Case Diagram
Data Trim
Data Regularization
Correlogram
Data Interpolation
t-test
Integration
Coefficient estimation
ForecastingResidual Generation
Residual CorrelogramPortmanteauTest
Plot of data
Plot of forecast
Administrator
Plot of correlogram
Content
20th July, 2006 Fareena Saqib NWF
Project Module
Content
20th July, 2006 Fareena Saqib NWF
Data Processing Module
Content
20th July, 2006 Fareena Saqib NWF
Data Cleaning Module Data files
Define format of the data
Trim operation
Regularization operation
Interpolation operation
Content
20th July, 2006 Fareena Saqib NWF
Data Processing
: Administrator
:Input
:InputOutputFile
:TrimData
:Regularization
:Interpolation
1: SelectType(type)2: SelectFormat(format)
3: openFile5: Extract()
9: Regularize()
4: SelectFile()
6: Extract(data,type,format)7: getTrimmedData()
8: returnTrimData(data)
10: Regularize(Trimdata)11: getRegularizeData()
12: returnRegularizeData(data)
13: Interpolate(Regularizedata)14: getInterpolateData()
15: returnInterpolateData(interpolatedata)
Content
20th July, 2006 Fareena Saqib NWF
Identification Module
Content
20th July, 2006 Fareena Saqib NWF
Identification Module Calculated autocorrelations.
Number of lags
Trend analysis of autocorrelation coefficients Correlogram
Content
20th July, 2006 Fareena Saqib NWF
Identification Module
: Administrator
:AutoCorrelati
:InputOutputTrimFile
:Input
:Plot Graph
12: plot()
:Integration
2: openFile()4: selectParameter(param)
5: paramData()8: Correlogram()
1: lags(lags)
9: getAutoCorrelation coefficient()
10: returnAutoCorrelation(data)
3: selectFile()
11: plotCorrelations(data)
6: Integration(data)
7: IntegratedData(data)
Content
20th July, 2006 Fareena Saqib NWF
Estimation Module
Content
20th July, 2006 Fareena Saqib NWF
Estimation Module Estimation Issues:
Random variable generation with normal distribution.
Estimation of the Model Estimation of the order Estimation of coefficient Estimation of the equation Testing the t-test of the coefficient
Content
20th July, 2006 Fareena Saqib NWF
Order Estimation
: Administrator
:Input
:Integrate
:OrderEstimate
:t-test
1: selectData()4: orderEstimate()
2: integrateData()
3: returnIntegratedData(indata)5: order(indata)8: returnAnalysisData(data)
6: testCoefficient(standardarror)
7: returnSignificance()
Content
20th July, 2006 Fareena Saqib NWF
Forecasting Module
Content
20th July, 2006 Fareena Saqib NWF
Forecasting Module Processed data with equal time intervals.
Estimation of order.
Formulate Equation.
Estimation of coefficients.
Forecast parameter values.
Plot Graph of forecasted values.
Content
20th July, 2006 Fareena Saqib NWF
Forecasting
: Administrator
:Input
:InputOutputFile
:EstimateCoefficient
:Forecast
:Plot
:CalculateDates:Residual
Estimation
11: forecast(forecastData)
1: orderAR(ar)2: orderMA(ma)
3: forcast()13: plotForecast()
4: forecast(data)
14: PlotForecast(dates,forecastdata)
7: Coefficient(ar,ma,data,rr)
8: returnCoefficient(c)
12: writeData(dates,forecastData)
9: futureDates(t)
10: returnDates(dates)
5: ResidualEstimate(ar,ma,data)
6: residuals(rr)
Content
20th July, 2006 Fareena Saqib NWF
Residual test Module
Content
20th July, 2006 Fareena Saqib NWF
Residual Module Graph of the residuals to check white noise
Check if the forecasting is valid or not.
Correlogram of residuals.
Portmanteau tests To test the Q-values
Content
20th July, 2006 Fareena Saqib NWF
Residual test
: Administrator
:Input
:ResidualPlot
:ResidualCorrelogram
:Portmanteau test
:AutoCorrelation
1: plotResidual()3: enterLags(l)
2: plot(residual)
4: rCorrelogram(rresidual)
7: returnCovarience(cov)
8: performTest(correlation)
9: sendAnalysis(analysis)
5: getCorrelation()
6: returnCorrelation(correlation)
Content
20th July, 2006 Fareena Saqib NWF
Visualization Module
Content
20th July, 2006 Fareena Saqib NWF
Visualization Module Visualization of steps carried by algorithm.
Visualization of the tool developed.
Plots of data at different processing stages.
Content
20th July, 2006 Fareena Saqib NWF
Demo: Snapshot of tool
Content
20th July, 2006 Fareena Saqib NWF
Results
Content
20th July, 2006 Fareena Saqib NWF
Content
20th July, 2006 Fareena Saqib NWF
Correlogram
Content
20th July, 2006 Fareena Saqib NWF
ARMA/ARIMA results
200
300
400
500
600
700
800
900
1000
1100
25 50 75 100 125 150 175 200
THMAX THMAXF
Content
20th July, 2006 Fareena Saqib NWF
Comparison Module
Content
20th July, 2006 Fareena Saqib NWF
Data plot of xtr
Content
20th July, 2006 Fareena Saqib NWF
Comparison of Results
0
50
100
150
200
250
300
350
400
450
500
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
HW
ARIMA
Content
20th July, 2006 Fareena Saqib NWF
Time Line
Content
20th July, 2006 Fareena Saqib NWF
ConclusionThe basic requirement of Network Weather Forecasting has been achieved by resorting to two prong efforts.
Firstly the technique of ARMA/ARIMA was followed and secondly an Algorithm was developed, both of which converged in dynamic Network data forecasting.
The methodology adopted was:
Available data on the subject was gathered and processed to be used as an input to the forecasting module.
After studying ARMA/ARIMA and ascertaining its suitability, algorithm was developed.
Based on the adopted approach and developed algorithm, experiments on forecasting were conducted employing the duly processed data.
The results obtained through different experiments were computed, compared and characteristics were plotted to come out with a fair idea of the final accomplishment.
Analysis of the results, their comparisons and other details were carried out before preparation of the report.
Documentation was undertaken to compile the project report.
Content
20th July, 2006 Fareena Saqib NWF
Research Accomplishments
Developed an algorithm for network weather forecasting.
A research paper on using ARIMA approach in network weather forecasting.
A Journal on results of different forecasting techniques on network data and their comparative analysis.
Content
20th July, 2006 Fareena Saqib NWF
Future Recommendations The forecasting carried out is based on a single approach which could
be explored for new dimensions.
The forecasting was carried out by employing three different tools of data collection which could be expanded to more numbers of tools in the future.
Efforts maybe initiated to apply new techniques like neural networks and others which are bound to come up in the fast developing field of information technology.
Forecasting of data should be made universal and the present form of retaining it on a single machine could be transformed as a web based tool serving all surfers of the web.
Content
20th July, 2006 Fareena Saqib NWF
Demo
Content
20th July, 2006 Fareena Saqib NWF
Thank You!
20th July, 2006 Fareena Saqib NWF
Appendix
20th July, 2006 Fareena Saqib NWF
Trim
: Administrator
:Input :InputOutputFile
:TrimData :TrimAbing :TrimIperf :TrimThrouhlay
1: SelectType(type)
2: SelectFormat(format)
3: openFile
4: SelectFile()
5: Extract(data,type,format)6: Extract(data,type,format)
7: [type=abing]Trim(data)
8: [type=iperf]Trim(data)
9: [type=throulay]Trim(data)
20th July, 2006 Fareena Saqib NWF
Regularization : Administrator
:Input :InputOutputTrimFile
:RegularizeData
:RegularizeAbing
:RegularizeIperf :RegularizeThrouhlay
1: SelectType(type)
2: openFile
4: Regularize(data,type)
3: SelectFile()
5: Regularize(data,type)6: [type=abing]Regularize(data)
7: [type=iperf]Regularize(data)
8: [type=throulay]Regularize(data)
20th July, 2006 Fareena Saqib NWF
Interpolation
: Administrator
:Input :InputOutputRegularizeFile
:RegularizeData
:InterpolateAbing
:InterpolateIperf :InterpolateThrouhlay
1: SelectType(type)
2: openFile
4: Interpolate(data,type)
3: SelectFile()
5: Interpolate(data,type)6: [type=abing]Interpolate(data)
7: [type=iperf]Interpolate(data)
8: [type=throulay]Interpolate(data)
20th July, 2006 Fareena Saqib NWF
Parameter estimation : Administrator
:Input :ParamGenerator
:InputOutputTrimFile
:DataFormatter
1: selectTool()
4: openFile()
6: selectParameter(param)
7: paramData()
2: generateParameterList
3: parameterList(param)
5: selectFile()
8: ExtractParam(param)
9: dataforForecast(data)
20th July, 2006 Fareena Saqib NWF
Identification Module : Administrator
:AutoCorrelation
:Plot Graph:Integration:InputOutputTrimFile
:Input:Input
1: lags(lags)
3: selectFile()2: openFile()
4: selectParameter(param)
5: paramData()
10: returnAutoCorrelation(data)
9: getAutoCorrelation coefficient()8: Correlogram()
11: plotCorrelations(data)
12: plot()
6: Integration(data)
7: IntegratedData(data)
20th July, 2006 Fareena Saqib NWF
Correlogram
: Administrator
:AutoCorrelation
:Plot Graph:Integration:InputOutputTrimFile
:Input:Input
1: lags(lags)
3: selectFile()2: openFile()
4: selectParameter(param)
5: paramData()
10: returnAutoCorrelation(data)
9: getAutoCorrelation coefficient()8: Correlogram()
11: plotCorrelations(data)
12: plot()
6: Integration(data)
7: IntegratedData(data)
20th July, 2006 Fareena Saqib NWF
Order Estimation
: Administrator:Input :Integrate :OrderEstimate :t-test
1: selectData()
4: orderEstimate()
2: integrateData()
5: order(indata)
3: returnIntegratedData(indata)
6: testCoefficient(standardarror)
7: returnSignificance()
8: returnAnalysisData(data)
20th July, 2006 Fareena Saqib NWF
Forecast : Administrator
:Input :InputOutputFile
:Plot:ResidualEstimation
:CalculateDates
:EstimateCoefficient
:Forecast
1: orderAR(ar)
2: orderMA(ma)
3: forcast()4: forecast(data)
7: Coefficient(ar,ma,data,rr)
8: returnCoefficient(c)
11: forecast(forecastData)
12: writeData(dates,forecastData)
9: futureDates(t)
10: returnDates(dates)
5: ResidualEstimate(ar,ma,data)
6: residuals(rr)
14: PlotForecast(dates,forecastdata)
13: plotForecast()
20th July, 2006 Fareena Saqib NWF
Residual tests : Administrator
:Input :InputOutputFile
:Plot:ResidualEstimation
:CalculateDates
:EstimateCoefficient
:Forecast
1: orderAR(ar)
2: orderMA(ma)
3: forcast()4: forecast(data)
7: Coefficient(ar,ma,data,rr)
8: returnCoefficient(c)
11: forecast(forecastData)
12: writeData(dates,forecastData)
9: futureDates(t)
10: returnDates(dates)
5: ResidualEstimate(ar,ma,data)
6: residuals(rr)
14: PlotForecast(dates,forecastdata)
13: plotForecast()
20th July, 2006 Fareena Saqib NWF
Trim
: Administrator
:Input
:InputOutputFile
:TrimData
:TrimAbing
:TrimIperf
:TrimThrouhlay
1: SelectType(type)2: SelectFormat(format)
3: openFile5: Extract(data,type,format)
4: SelectFile()
6: Extract(data,type,format)
7: [type=abing]Trim(data)
8: [type=iperf]Trim(data)9: [type=throulay]Trim(data)
20th July, 2006 Fareena Saqib NWF
Regularization
: Administrator
:Input :InputOutputTrimFile
:RegularizeData
:RegularizeAbing
:RegularizeIperf
:RegularizeThrouhlay
1: SelectType(type)2: openFile
4: Regularize(data,type)
3: SelectFile()
5: Regularize(data,type)
6: [type=abing]Regularize(data)
7: [type=iperf]Regularize(data)
8: [type=throulay]Regularize(data)
20th July, 2006 Fareena Saqib NWF
Interpolate
: Administrator
:Input
:InputOutputRegularizeFile
:RegularizeData
:InterpolateAbing
:InterpolateIperf
:InterpolateThrouhlay
1: SelectType(type)2: openFile
4: Interpolate(data,type)
3: SelectFile()
5: Interpolate(data,type)
6: [type=abing]Interpolate(data)
7: [type=iperf]Interpolate(data)
8: [type=throulay]Interpolate(data)
20th July, 2006 Fareena Saqib NWF
Data Processing
: Administrator
:Input
:InputOutputFile
:TrimData
:Regularization
:Interpolation
1: SelectType(type)2: SelectFormat(format)
3: openFile5: Extract()
9: Regularize()
4: SelectFile()
6: Extract(data,type,format)7: getTrimmedData()
8: returnTrimData(data)
10: Regularize(Trimdata)11: getRegularizeData()
12: returnRegularizeData(data)
13: Interpolate(Regularizedata)14: getInterpolateData()
15: returnInterpolateData(interpolatedata)
20th July, 2006 Fareena Saqib NWF
Parameter selection
: Administrator
:Input
:ParamGenerator
:InputOutput
:DataFormatter
1: selectTool()4: openFile()
6: selectParameter(param)7: paramData()
2: generateParameterList
3: parameterList(param)
5: selectFile()
8: ExtractParam(param)
9: dataforForecast(data)
20th July, 2006 Fareena Saqib NWF
Order estimation
: Administrator
:Input
:Integrate
:OrderEstimate
:t-test
1: selectData()4: orderEstimate()
2: integrateData()
3: returnIntegratedData(indata)5: order(indata)8: returnAnalysisData(data)
6: testCoefficient(standardarror)
7: returnSignificance()
20th July, 2006 Fareena Saqib NWF
Forecasting
: Administrator
:Input
:InputOutputFile
:EstimateCoefficient
:Forecast
:Plot
:CalculateDates:Residual
Estimation
11: forecast(forecastData)
1: orderAR(ar)2: orderMA(ma)
3: forcast()13: plotForecast()
4: forecast(data)
14: PlotForecast(dates,forecastdata)
7: Coefficient(ar,ma,data,rr)
8: returnCoefficient(c)
12: writeData(dates,forecastData)
9: futureDates(t)
10: returnDates(dates)
5: ResidualEstimate(ar,ma,data)
6: residuals(rr)
20th July, 2006 Fareena Saqib NWF
Residual test
: Administrator
:Input
:ResidualPlot
:ResidualCorrelogram
:Portmanteau test
:AutoCorrelation
1: plotResidual()3: enterLags(l)
2: plot(residual)
4: rCorrelogram(rresidual)
7: returnCovarience(cov)
8: performTest(correlation)
9: sendAnalysis(analysis)
5: getCorrelation()
6: returnCorrelation(correlation)
20th July, 2006 Fareena Saqib NWF
Thank you!!!