INDERPRASTHA ENGINEERING COLLEGE

CIVIL ENGINEERING DEPARTMENT

PROJECT PRESENTATION
ON

FORECASTING AIR QUALITY PARAMETERS

USING LINEAR & NONLINEAR MODELS
MENTOR GROUP NO- SUBMITTED BY
Prof: Sheelu Verghese 08 Mian Umeria Masood(1803000903)
Asst Professor, CED Aryan Pandey(1703000006)
 FORECASTING AIR QUALITY
PARAMETERS USING LINEAR AND
NONLINEAR MODELS
 01

PRESENTATION OUTLINE
S NO. CONTENT NAME
1. INTRODUCTION
2. LITERATURE REVIEW
3. PROBLEM STATEMENTS
4. OBJECTIVES OF STUDY
5. METHODOLOGY
6. SCOPE OF STUDY
7. ADVANTAGES AND DISADVANTAGES
8. REFERENCES
 INTRODUCTION 02

The pollution concentrations for criteria pollutants in most Indian cities exceeds the national
ambient air quality limit standards.
Variation in air pollution concentrations maybe partly due to emissions and chemical
transformations but the role of meteorological conditions must also be investigated.
Extreme air pollution episodes affect millions of people in densely populated regions and it
is responsible for health problems like asthma, bronchitis, lung cancers etc. in urban areas
and hence is a cause of concern.
A study of WHO highlights that deaths from air pollution in India will rise from 1mn in
2015 to 1.7mn deaths annually by 2050.
Meteorological parameters are known to have a strong influence on concentrations of
pollutants in the atmosphere
The pollution concentrations in urban areas is a function of Temperature, Pressure, Wind
Speed, Relative Humidity etc.
NATIONS WITH THE HIGHEST AND LOWEST LEVELS OF AIR POLLUTION 03

Image SOURCE : 
WEF
 05
CONSEQUENCES OF AIR POLLUTION
 06

LITERATURE REVIEW
AUTHOR & TITLE OBJECTIVE STUDY METHODOLOGY 07
YEAR AREA
The relationship Deriving the 68 major Data Analysis using
b/w PM 2.5 and relationship in cities MLR & ANN tool
meteorological CHINA
factors in china:
Yang et al. Seasonal &
2017 Regional
Variations

CONCLUSION

RH is positively related with PM 2.5 concentrations in North China, but negatively

correlated with PM 2.5 is stronger in South China and other parts. This positive
correlation b/w RH & PM 2.5 is stronger in winter & spring
AUTHOR TITLE OBJECTIVE STUDY METHODOLOGY 08
& YEAR AREA
Statistical Relationship b/w Elazig; Linear regression was
Analysis of monitored air Turkey used to find the
meteorological pollution relationship b/w the
Akpinar et factors & air concentration variables ( the value of
al. pollution at such as SO2 and independent variable
2009 winter months the Total was up to 1)
Suspended
Particles(TSP)
data and
meteorological
factors

CONCLUSION

There is a moderate & weak level of relationship b/w the air pollutant concentrations &
the meteorological factors in Elazig city
AUTHOR & TITLE OBJECTIVE STUDY METHODOLO09
YEAR AREA GY
Reversing To generate and test the air 5 locations Regression
Urban air quality forecasting models models are being
quality and to suggest the CI based created based on
Karatzas et forecasting ensemble oriented models ensemble
al. that are able to predict as modeling
2011 much information as principle by using
possible from atmospheric ANN tool
quality data of low
dimensionality

CONCLUSION

Overall result suggest that the specific modelling approach can support the provision of the
quality forecasts on the basis of limited features space dimensionally & by employing
simple regression models
AUTHOR & TITLE OBJECTIVE STUDY AREA METHODOLO10
YEAR GY

Impacts of Prediction Industrial & IBM SPSS21

Ozdemir and meteorological capabilities of Urban regions of statistical
Taner parameters on MLR & ANN kocaeli, Turkey program for
          2013 PM 10 were to be MLR analysis &
investigated for ANN three
layers were used

CONCLUSION

• This study showed that the ANN model system are more efficient for prediction &
sensitivity of data sets.
• ANN showed correlation coefficients R2 values as 0.74 & 0.36 for Industrial & Urban
regions.
• MLR showed correlation coefficients R2 values as 0.87 & 0.49 for Industrial & Urban
regions.
AUTHOR & TITLE OBJECTIVE STUDY METHODLOGY 11
YEAR AREA
Abdullah et al. Forecasting Precise Peninsula In Linear model ;
2015 Particulate forecasting of East MLR was used and in
matter Algorithms were Africa Non Linear model ;
concentration urgently Multiple layer
using Linear and required to perception & Radial
Non-Linear determine PM basis function was
models 10 status used

CONCLUSION

The concluded study used data of 14years from (2000-2014)and concluded that non –linear
algorithm is better in forecasting the next day’s concentrations with Radial Basis
Function(RBF) giving best predicted value.
AUTHOR & TITLE OBJECTIVE STUDY METHODOLOGY 12
YEAR AREA
Evaluation of To predict the East-Cost Full model of ANN &
Yousuf et al. ANN & MLR future haze of MLR with 14 variables
2018 over PM10 occurrences Malaysia were used as input.
during haze and Sensitivity Analysis
non haze was coupled with
conditions models to recognize the
pollutant relationship
PM10

CONCLUSION

ANN performed better in both models development compared to MLR either in ‘r’
evaluation or in determining RMSE & SSE values. In scientific perspective, ANN was
partial than MLR and predicted that meteorological parameters were dominant in both hazy
and non hazy conditions
 13
CONCLUDING REMARKS

This type of correlation analysis can be carried out to predict the relation between
the air pollution and meteorological parameters.

Correlation Analysis establishes the strength of association b/w meteorological variables

and pollutant concentrations.

MLR & ANN both can be used to predict future concentrations but ANN is a better tool.

ANN takes into concentration the non-linear relationship between the various inputs and
assigns weights and bias accordingly.

Meteorology influences pollution built up differently for a urban & an inland urban city .
 14
PROBLEM STATEMENT
Meteorological parameters influences
Correlation Study is significant to
pollution concentrations differently. It
map the air pollution-
exhibits regional and seasonal
meteorology association.
variations.

Statistical and Artificial

The forecasted results of MLR &
intelligence forecasting methods
ANN analysis needs to be validated
based on data of the air pollution
for New Delhi and West Bengal for
is a useful tool in predicting
effective implementation.
pollution concentrations.
 15
OBJECTIVE OF STUDY
1 To establish correlation between meteorological parameters and particulate
pollutant concentrations.

To develop a forecasting model based on meteorology-air quality

2 association using MULTIPLE LINEAR REGRESSION(MLR) &
ARTIFICIAL NEURAL NETWORKS(ANN) APPROACHES

3 To check which of the mentioned model is more accurate than the other.

4 To compare the results for the two location( Delhi and West Bengal)
 16
METHODOLOGY
Compile the daily average pollutant concentrations and
1 meteorology variables from CPCB website

Carry out data normalization to bring it in the scale 0 to 1.

2

Carry out MLR Analysis in MS EXCEL & ANN

3 modelling in MATLAB.
Validate the results obtained for future prediction using
4 Statistical Indices.
Compare the results of MLR & ANN for the two
5 locations( Anand Vihar and Asansol)

Give suitable recommendations.

6
 17

CORRELATION

Value of ‘r’
Correlation in Excel
 18

REGRESSION

Regression Eqn.(multiple variable)

 19
SCOPE OF STUDY
In the study, relationship between monitored air pollution and meteorological factors are to
be investigated from January 2019 – December 2019.
The meteorological factors to be considered in this study are temperature, wind speed,
relative humidity, pressure etc. from which conclusions as to the fore coming weather are
drawn. Also the pollutants considered are PM2.5 AND PM10.

Pollution has direct impact on health, Accurate forecasting of pollution concentration helps
people and govt to plan ahead, decreasing the effects in health and cost associated.
Regulatory authorities can utilize air quality forecast to implement effective pollution
control measures.

The study enables the comparison of performance of regression models and ANN tools
in forecasting.
 20

ADVANTAGES AND APLICATIONS

To inform the general public To forecast pollution

about the expected pollution concentrations for timely
levels. preventive measures.

To take strategic decisions, such

To protect the health of sensitive as shutting down of maximum
people (like infants, aged polluting industries which aren’t
people) by providing advance following norms provided by
warnings. the government.
WHERE DID DATA COME FROM ? 21

We click on Advance
Search..
 22

We enter all the required

fields

And, after this our

Step 01 is completed! 
 22
DATA COLLECTED

Asansol, West Bengal Anand Vihar, New Delhi

 23
02 – DATA NORMALIZATION!

Normalization is useful when your data has varying scales. The goal of normalization is to
change the values of numeric columns in the dataset to a common scale, without distorting
differences in the ranges of values.

Formula for normalization in Stats Formula for normalization in Excel

 24

NON -NORMALIZED NORMALIZED

METEROLOGICAL METEROLOGICAL
PARAMETERS PARAMETERS
 coefficient of Correlation(r).. 25
STEP 03?

Method 1 – Using CORREL

Method 2 – Using DATA ANALYSIS TOOLPAK 26

1.

2.

3.
 27

4. 5.

Giving Output range

(where to display result)

Giving Input Range, here

 28
5. Result 

Results obtained from both methods are same but Method 1 is easy, why ?

In Method 2, our columns must be adjacent.

 29
Project Correlation Result! 

Asansol, W.B Anand Vihar, Delhi

 Correlation of A&B = Correlation of B&A

 30

MULTIPLE LINEAR
STEP 03 REGRESSION!

How various Independent variables affect the dependent variables.

 31
How to do REGRESSION !
 Imp. Terms . 32

What Regression tells us ? Overall Regression can be

- Explains 2 Things! explained by 3 things -
1. R^2
1. Overall fit of the model. 2. Significance F &
2. Individual Predictors. 3. P-values

R^2? P-Values ?
Significance F ?
- How much of - Individual
- always < 0.05.
dependent variable is
explained by independent - Test is Significant. Predictor .
variables.
Regression Results - 33

DELHI

PM2.5 = 3.57*RH+(14.7*WS)+(7.39*WD)+(-18.58*SR)+(1.64*BP)+(-1.91*AT) +
 34

PM10 = 1089*(RH)+(-29.11*WS)+(-8.1*Wd)+(-70.21*SR)+(5.89*BP)+(-107.76*AT) +
1398
 35

WEST BENGAL

PM2.5=  ( -19.2*RH ) + ( -5.4*WS) + (-23.3*WD) + (-1.14*SR) + (19.98*BP) + (-15.63*AT ) + 215.14

 36

PM10 =  ( -2.57*RH ) + ( -8.4*WS) + (3.3*WD) + (2.02*SR) + (8.98*BP) + (4.3*AT ) + 393.06

 RELATIVE ERROR CALCULATION 

RELATIVE ERROR FORMULA

PERCENTAGE RELATIVE ERROR

CALCULATION OF RELATIVE ERROR 37

Anand Vihar,
DELHI

Asansol, W.B.
 38
MATLAB?

• Programming platform.
• Designed specifically for engineers and scientists to analyze and design systems and
products that transform our world.

FEATURES - 
• Machine Learning ,
• Deep Learning ,
• Predictive Maintenance ,
• Test and Measurement ,
• Robotics ,
• Image Processing and Computer Vision ,
• Machine Learning etc.
 39
ANN MODELING - MATLAB

MATLAB GUI
DOING MODELING FOR PM2.5 - ANAND VIHAR 40

• Creating NEW WORKSPACE

• Importing Input, Target data.
• Transposing all the imported data.

Input Data Target data

 41

Command Window
Importing Data to DATA MANAGER
 42
ANALOGY B/W BRAIN & NN

HUMAN BRAIN – NN – Association of

Association of Neurons.  Weighted Neurons. 
 43
LAYERS OF NN 
 44
CREATING A NETWORK

TRIAL - 1
 45

• Create New Network,

• Change no. Of Neurons,
• Change Performance function,
• Change Training Function,
• Reassign Weights,

DO VARIOUS COMBINATIONS /
DO HIT TRIALS.

Result of TRIAL 1
 46

Result of TRIAL 2
 47

RESULT - TRIAL 03
 48

TRIAL 04
 49

TRIAL 05 TRIAL 06
 50

Reverted Weights Changed Input Ranges

Created 'n' Network and Bias(es)
Combinations
 51

&
Edited Weights

Changed Epochs
 Final Result 52

• ACTIVATION FUNCTION - TRAINLM

• Training Algorithm 
-  LEVENBERG MARQUARDT     
 ALGO.
• NEURONS No - 20
• EPOCHS No - 10

After n attempts value of R = 0.96

 53

Either EXPORT or SAVE values

 54

TRAIN >  OUPUT STORAGE/ OUPUT EXPORT > WORKSPACE

Output WORKSPACE PREDICTED VALUES

 55

Similarly It will be done for PM10 & Pollutants of Asansol, WEST

BENGAL .

Finding RELATIVE ERROR b/w PREDICTED VAL. & OB. VALUES.

 56
RELATIVE ERROR

ANAND VIHAR

ASANSOL
 CONCLUSION - 57
•       The Relative error percentage for PM2.5 of Anand Vihar was found to be 5.81 using MS-EXCEL and 1.57 using
ANN Tool of MATLAB.

•      The Relative error percentage for PM10 of Anand Vihar was found to be 5.69 using MS-EXCEL and 1.33 using
ANN Tool of MATLAB.

•     The Relative error percentage for PM2.5 of Asansol was found to be 7.73 using MS-EXCEL and 2.61 using ANN 
 Tool of MATLAB.

•     The Relative error percentage for PM10 of Asansol was found to be 4.17 using MS-EXCEL and 2.41 using ANN
Tool of MATLAB.

• Both the models are effective in prediction analysis but ANN performed much better and having higher accuracy.

• Models can be applied to simulate pollutants concentrations during episodic crisis so that better management steps
could be taken in advance preventing harm of larger population.

• We can identify major pollutants sources and take appropriate control measures.

PREVENTION IS BETTER THAN CURE.

 REFERENCES 58
• Abdul-Wahab, S. A., Bakheit, C. S., and  Al-Alawi, S. M. 2005. Principal component  and multiple
regression analysis in modelling  of ground level ozone and factors affecting  its concentrations.
Environmental Modelling and Software. 20:1263–1271.
• Ash, K. D. & Matyas, C. J. 2012. The  influences of ENSO and the  subtropical Indian Ocean
Dipole on  tropical cyclone trajectories in the  southwestern Indian Ocean. International Journal
of Climatology. 32(1).
• Alencar, A., Nepstad, D. & Vera-Diaz, M. d-C. 2006. Forest understory fire in  the Brazilian
Amazon in ENSO and  non-ENSO years: Area burned and  committed carbon emissions. Earth
Interactions. 10(6), 1–17.
• Akpinar, E.K., Akpinar, S. & Öztop, H.F. 2006. Effects of meteorological parameters on air 
pollutant concentrations in Elazig. Turkey. Int. J.  Green Energy. 3, 407−21.
• Hu, X.; Waller, L.A.; Al-Hamdan, M.Z.; Crosson, W.L.; Estes, M.G.; Estes, S.M.;  Quattrochi,
D.A.; Sarnat, J.A.; Liu, Y.  Estimating ground-level PM2.5  concentrations in the southeastern US
using  geographically. 456-854.
• Kovac-Andric, E.; Brana, J.;  Gvozdic, V. 2009. Impact of  meteorological factors on ozone 
• Yang, F. S., Nuhoglu, A. & Bayraktar, H.  2005. Impacts of some meteorological  parameters on
SO2 and TSP concentrations  in Erzurum, Turkey, Chemosphere 59,  1633−1642.