b. matić et al., thailand
TRANSCRIPT
-
8/17/2019 B. Matić Et Al., Thailand
1/4
The 7th International Conference on Engineering and TechnologyICET-2015, Phuket, June 19-20, 2015
Prince of Songkla University, Faculty of EngineeringHat Yai, Songkhla, Thailand 90112
Abstract: Many regression models have been developedto predict the temperature of an asphalt concrete (AC)
layer using air temperature with some other input
parameters. Some of these models are aged and cannot
be applied to diverse site locations with accuracy. This
paper presents models to predict surface pavement
temperature depending on the days of the year using
neural networks. Four annual periods are defined and
new models are formulated for each period. Models are
formed on the basis of data gathered on the territory of
the Republic of Serbia and are valid for that territory.
Key Words: pavement, temperature, model, predicting, artificial neural networks (ANN)
1. INTRODUCTION
Authors all around the world have provided models
to predict pavement temperature both on local and on
global level.
Several models predicting the pavement temperaturehave been analysed, and it has been determined that the
best predicting models are the ones formed using
databases for the territory where they are intended to
predict pavement temperature.
Pavement temperature can be influenced, apart from
latitude and other factors, by seasons as well, and by
time of a day.Until today, many authors have formed models using
statistical data analysis to predict the surface pavement
temperature.
Table 1. An overview of models for predicting
pavement temperatures using regression analysis
[1,2,3,4,5,6]
AuthorThe publishing
year
Statistical
data
analysis
Strategic Highway
Research Program
(SHRP)
1987 Yes
Mohseni A. 1998 Yes
Lukanen et al. 1998 Yes
Bosscher et al. 1998 YesOvik et al. 1999 Yes
Park et al. 2001 Yes
Marshall et al. 2001 Yes
Denneman E. 2007 Yes
Abo-Hashema, M. (2013) discusses the feasibility of
applying artificial neural network (ANN) technology in
predicting the AC layer temperature. In his paper, the
neural network is trained and tested usingNeuroSolutions 5.0 software through the actual field data
obtained from Long-Term Pavement Performance
(LTPP), Seasonal Monitoring Program (SMP) -
DataPave3.0 software.Results indicate that the developed ANN-based
pavement temperature prediction models can be used inpredicting AC layer temperature with high accuracy in
comparison to the measured values.
This outcome is considered crucial to the pavement
design, especially followed by the second ANN model
where some input parameters may not be available [7].
B. Matić et al., (2014) formulate a new model forpredicting pavement temperatures at specified depth
using neural networks, depending on the surface
pavement temperature and depth.
Based on the correlation coefficient, standard model
deviation and mean absolute error (MAE), it could be
concluded that these models could be utilized with theadequate accuracy for predicting maximal and minimal
pavement temperature at depth; yet, the model for
predicting minimal pavement temperature at certain
depth provides a somewhat more accurate result.
Also, it can be concluded that the models formed byANN predict pavement temperatures with higher
accuracy than models formed by the regression analysis
[8].
Seasonal Asphalt Concrete Surface Pavement
Temperature Models Using Neural Networks
Bojan Matić *, Vlastimir Radonjanin , Mirjana Malešev , Nebojša Radović , Siniša
Sremac1 1University of Novi Sad, Faculty of Technical Sciences, Novi Sad, Serbia
*Email of corresponding author: [email protected]
-
8/17/2019 B. Matić Et Al., Thailand
2/4
2. EXPERIMENTAL PART
2.1. Instrumentation and data
Data used for analyzing and forming the model for
pavement temperature prediction were taken from thepilot project by the public company (JP)“Putevi Srbije”
and the Government of Sweden.
Surface pavement temperature, as well as airconditions, were monitored using modern meteorological
stations (Road Meteorological Information System) on
six locations during the part of 2010, the entire 2011 andthe part of 2012 for winter road maintenance.
The plan was to monitor temperature in winter
months only.
However, following the insistence and demand by the
Faculty of Technical Sciences in Novi Sad, Serbia to
monitor temperature during the entire year for the
purpose of scientific research, JP “Putevi Srbije” grantedthe monitoring in other months as well.
The acquisition of all meteorological data in the base,
including temperature, was performed every 30 minutes.
Following values were measured: air temperature, air
humidity, dew point temperature, type of precipitation,
surface pavement temperature, and max wind speed [9].
3. RESULTS AND DISCUSSION
3.1. Methodology and anticipated results
The model developed with the objective of predicting
surface pavement temperature is based on the neuralnetworks analysis. Models are formed to predict
maximum and minimum surface pavement temperatures,
depending primarily on the maximum and minimum air
temperature and depending on the days of the year.
3.2. Model validation
The paper formulates new models for predicting
maximum and minimum pavement surface temperatures
using ANN, in dependence on the ambient air
temperature.
The former predicts maximum surface pavement
temperatures based on maximum air temperature, as seen
in Table 2. Меаn Average Error (МАЕ) for this model is
3.82937757 oC.
The latter predicts minimum surface pavementtemperatures based on minimum air temperature, as
observed in Table 3. Меаn Average Error (МАЕ) for this
model is 1.40998219oC.
Fig. 1. Road meteorological stations [9]
Table 2. Model predicting maximum surface pavement temperature based on maximum air temperature by ANNIndex Net.
nameTraining
perf.Testperf.
Trainingerror
Testerror
Trainingalgorithm
Errorfunction
Hiddenactivation
Outputactivation
1 MLP 1-5-1 0.944463 0.815718 0.004626 0.015968 BFGS 111 SOS Logistic Exponential
Table 3. Model predicting minimum surface pavement temperature based on minimum air temperature by ANNIndex Net.
name
Training
perf.
Test
perf.
Training
error
Test
error
Training
algorithm
Error
function
Hidden
activation
Output
activation1 MLP 1-5-1 0.974493 0.972041 0.001555 0.001618 BFGS 44 SOS Logistic Tanh
Table 4. Table overview of MAE of models predicting max and min surface pavement temperature by ANN
Меаn Average Error, МАЕ,oC
maxTk minTk
3.82937757 1.40998219
Based on previously completed research and
developed models using regression analysis, it is
concluded that models that predict the minimum and
maximum surface pavement temperature provide better
results in comparison to the models using ANN [10].
Furthermore, the paper formulates the models for
predicting surface pavement temperature for diverse
periods during a year based on air temperature, day of
the year and humidity by ANN. The Меаn Average
Errors (МАЕ
) of these models are shown in Table 5.
-
8/17/2019 B. Matić Et Al., Thailand
3/4
Table 5. Table overview of models predicting surface pavement temperature for diverse periods during a year by ANN
Independent Variables
Меаn Average Error, МАЕ, oC
January-
МаrchApril-Jun
July-
September
October-
December
Air temperature, day of year 0.894487 2.226435 1.690075 0.869849
Air temperature, day of year,
humidity1.033865 2.470103 1.683501 0.852367
If we use humidity as one of the input variables in the
model for predicting the surface pavement temperature,
the МАЕ will not be less for all models. For some
models MAE will be increased.
Accordingly, it can be concluded that the model for
predicting surface pavement temperature using ANN for
the first two periods of the year (from January to March
and from April to June) is better without humidity as aninput variable (i.e. model where the input variables are
air temperature and day of the year). For the other two
periods (from July to September and October to
December) a better model is with humidity as an input
variable.
The tables below present the ANN models that
predict seasonal asphalt concrete surface pavement
temperature with the highest accuracy using ANN for
four different periods of the year (Tables 6, 7, 8, 9).
Table 6. Model predicting surface pavement temperature by ANN based on air temperature and day of the year for
period January- Маrch Index Net.
nameTraining
perf.Testperf.
Trainingerror
Testerror
Trainingalgorithm
Errorfunction
Hiddenactivation
Outputactivation
1 RBF 2-63-1 0.895316 0.889872 0.002 0.0021 RBFT SOS Gaussian Identity
Table 7. Model predicting surface pavement temperature by ANN based on air temperature and day of the year for
period April-Jun
Index Net.name
Trainingperf.
Testperf.
Trainingerror
Testerror
Trainingalgorithm
Errorfunction
Hiddenactivation
Outputactivation
1 MLP 3-6-1 0.946513 0.964050 0.002701 0.003905 BFGS 47 SOS Logistic Logistic
Table 8. Model predicting surface pavement temperature by ANN based on air temperature and day of the year for
period July-September Index Net.
nameTraining
perf.Testperf.
Trainingerror
Testerror
Trainingalgorithm
Errorfunction
Hiddenactivation
Outputactivation
1 MLP 3-3-1 0.961315 0.955025 0.00 0.002 BFGS 50 SOS Tanh Identity
Table 9. Model predicting surface pavement temperature by ANN based on air temperature and day of the year for
period October-December Index Net.
nameTraining
perf.Testperf.
Trainingerror
Testerror
Trainingalgorithm
Errorfunction
Hiddenactivation
Outputactivation
1 MLP 3-4-1 0.978898 0.972124 0.000711 0.000916 BFGS 70 SOS Exponential Logistic
4. CONCLUSION
The paper formulates new models for predicting
minimum and maximum pavement surface temperaturesusing ANN, in dependence on the ambient air
temperature.
Also, the paper formulates new models for predicting
day surface pavement temperature, including seasonal
influences by ANN.
Furthermore, model validation has been
conducted.Based on the mean absolute error (MAE) and
standard deviation of error (SDE) between measured and
predicted pavement temperatures, it can be concluded
that the models formed with two variables (air
temperature and day of the year) by ANN predict
pavement temperatures for period January-March and
April-Jun with better accuracy than the models formed
by regression analysis.
Likewise, it can be concluded that the models formed
with three variables (air temperature, day of the year and
humidity) by ANN predict pavement temperatures for
period July-September and October-December with
higher accuracy than the models formed by twovariables.
-
8/17/2019 B. Matić Et Al., Thailand
4/4
Table 10. Table overview of models predicting surface pavement temperature for diverse periods during a year by
regression analysis and ANN
Independent
VariablesType of
analysis
Меаn Average Error, МАЕ, oC
January-
МаrchApril-Jun
July-
September
October-
December
Air temperature,
day of year
Regression
analysis [3]1.284492 2.799148 2.158237 1.123356
ANN 0.894487 2.226435 1.690075 0.869849
Air temperature,
day of year,
humidity
Regression
analysis [3]1.283519 2.613486 2.162538 1.082358
ANN 1.033865 2.470103 1.683501 0.852367
AcknowledgmentThe authors acknowledge the support of the research
project TR 36017, funded by the Ministry of Science andTechnological Development of Serbia.
5. REFERENCES
[1] T. Kennedy, G. Huber, E. Harrigan, R. Cominsky, C.
Hughes, H. V. Quintus, J. Moulthrop, “Superior
Performing Asphalt Pavements (Superpave): The
product of the SHRP Asphalt Research Program”,
National Research Council, Washington, DC, (1994).
[2] A. Mohseni, M. Symons, “Effect of Improved LTPP
AC Pavement Temperature Models on SuperPave
Performance Grades”, Proceedings of 77th Annual
Meeting, Transportation Research Board,
Washington, DC, (1998b).
[3] E. O. Lukanen, H. Chunhua, E.L. Skok,
“Probabilistic Method of Asphalt Binder Selection
Based on Pavement Temperature”, Transportation
Research Record, Transportation Research Board,
1609 (1998), 12-20.
[4] P. J. Bosscher, H.U Bahia, S. Thomas, J.S. Russel,
“Relationship Between Pavement Temperature and
Weather Data: Wisconsin Field Study to Verify
SuperPave Algorithm”, Transportation Research
Record, 1609 (1998), 1-11.
[5]C. Marshall, R. Meier, M. Welch, “Seasonal
Temperature Effects on Flexible Pavements inTennessee”, Transportation Research Record, 1764
(2001), 89-96.
[6] E. Denneman, “The application of locally developed
pavement temperature prediction Algorithms in
Performance Grade (PG) Binder Selection”, The
Challenges of Implementing policy-SATC 2007: The
26th Annual Southern African Transport Conference
and Exhibition, Pretoria, South Africa, (2007), p.10.
[7] Abo-Hashema M., “Modeling Pavement TemperaturePrediction Using Artificial Neural Networks”.
Airfield and Highway Pavement, 2013, pp. 490-505.
[8] B. Matić, D. Matić, Đ. Ćosić, S. Sremac, G. Tepić, P.
Ranitović, “A model for the pavement temperature
prediction at specified depth”, Metallurgy, 52 (2013)
4, 505-508.
[9]B. Matić, “Seasonal asphalt concrete surface
pavement temperature models”, Contemporary Civil
Engineering Practice, 2013, pp. 243-258.
[10] B. Matić et al., “Development and evaluation of the
model for the surface payment temperature
prediction”, Journal “Metalurgija”, Croatian
Metallurgical Society, Zagreb, Croatia, 51(2012)3,pp. 329-332, ISSN: 0543-5846.
.