GREECOPE: Green Computing with Piezoelectric Effect
Meghana Kshirsagar
1 a
, Rutuja Lahoti
, Tanishq More
and Conor Ryan
1 b
Biocomputing and Developmental Systems Group, Lero, University of Limerick, Ireland
Department of Information Technology, Government College of Engineering, Aurangabad, India
Keywords: Intelligent Transportation, Piezoelectric Effect, Smart Cities, Road Transport.
Abstract: The growing interest in the search and use of alternative resources for renewable energy can lead the future
towards substantially decreasing carbon footprint and reduce the effects of global warming. The proposed
research explores the possibility of harnessing piezoelectric energy from the environment of moving vehicles
on road. Although the technology is still immature, it has the advantage of having zero carbon footprints thus
making it ideal to investigate the potential for green energy generation. The main objective is to develop
regression models that can estimate energy generated from vehicular traffic. Energy is generated when force
is applied to piezoelectric transducers which depend on significant factors such as the number of piezoelectric
transducers and their arrangement, load applied and frequency. We design Support Vector Machine (SVM)
and Generalised Linear Model (GLM) for predicting energy. The best features for training the model were
selected by incorporating feature selection techniques such as Pearson’s correlation coefficient and Mutual
Information Statistics. The experimental setup makes use of simulated data which takes into account vehicle
count of different vehicles with and without load. The accuracy achieved from SVM and GLM are 99.6% and
99.7% respectively. The energy savings achieved by making use of generated piezoelectric energy is
discussed with a sample scenario of Motorway50 of Dublin, the Irish Capital city. Through this work, we
propose to investigate deeper into the feasibility towards cost-effectiveness by utilizing energy which is
wasted by human and vehicular locomotion.
Basic human needs largely rely upon non-renewable
sources (Reeker, 2014) of energy which will become
exhausted in the future. Thus, there's a major
imperative to seek out some way through which we
are able to generate energy from the available
resources. This can cut back the use of non-renewable
energy sources to a good extent and may help to
create a better future. Solar power, windmills and
geothermal energy are some examples of renewable
energy (Davies, 2017). Petrol and diesel are non-
renewable resources and contribute significantly to
the effects of heat and climate impact over time as
travel has become one of the most important
requirements of our lifestyle, which is made possible
by the use of a variety of vehicles. Vehicles use fossil
fuels, which release large amounts of carbon and
greenhouse gas. Electric vehicles are becoming
increasingly more popular and their current limited
use presently is expected to scale up in the coming
decade. Hence with the anticipated growing trend for
electric vehicles the use of electricity as its fuel would
substantially increase. Fortunately, the piezoelectric
effect can utilize waste locomotion energy and
vehicle kinetic energy and convert it into electricity.
This can be achieved through the use of piezoelectric
sensors that can generate electrical energy when
pressure is applied to it. Therefore, energy can be
built up through the movement of road vehicles and
people (Borikar, 2017).
In this work a comprehensive model to predict the
piezoelectric energy output based on historical data is
discussed. We explore the use of Artificial
intelligence-based methods like Support Vector
Machine (SVM) (Kaur, 2017) and Generalized
Linear Model (GLM) (McCullagh and Nelder, 1989)
for the estimation of electricity generation through
piezoelectric effect from moving road vehicles is
discussed. Artificial intelligence-based methods have
the twofold advantage of being flexible and capable
of dealing with non-linearity.
Kshirsagar, M., Lahoti, R., More, T. and Ryan, C.
GREECOPE: Green Computing with Piezoelectric Effect.
DOI: 10.5220/0010445801640171
In Proceedings of the 10th International Conference on Smart Cities and Green ICT Systems (SMARTGREENS 2021), pages 164-171
ISBN: 978-989-758-512-8
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
This section will discuss the characteristics of
piezoelectric material, their application and the
formulation for calculation of generation of energy.
2.1 Piezoelectric Materials,
Characteristics & Formula
Currently the most extensively used piezoelectric
material is PZT (Pb- Lead, Zr- Zirconium, Ti-
titanium). The working process of the piezoelectric
system when placed beneath the road is explained in
(Kour and Charif, 2016).
The proposed system is completely based on the
sound theoretical concepts of Piezoelectric effect,
which is the generation of electric charge when
subjected to pressure. Materials which exhibit
piezoelectric effect are known as piezoelectric
Certain factors affect the amount of power
generated by incorporating piezoelectric plates
beneath the road surfaces, such as vehicle speed,
vehicle mass, traffic flow, dimension of road on
which piezoelectric plates are embedded and the
frequency settings depending upon locations. Speed
and weight of a vehicle is directly proportional to
energy generated.
Below formulas are useful while preparing the
dataset mentioned in 3.2 and regression model.
Let the total no of vehicles be T
and individual
vehicle count of car, bus, truck, motorcycle is T1, T2,
T3 and T4 respectively. Consider the average mass of
total no car, bus truck and motorcycle as M1, M2, M3
and M4 respectively.
Rolling resistance Fr of the wheels is calculated by
     
Where N
is the normal force acting on the surface
Cr - coefficient of rolling friction (0.03-0.15)
Mi = Mass of particular vehicle class, where
i (1,4)
g = Gravity
The rolling resistance of the tire can be overcome by
the power given by
  
Where v = speed of vehicle.
Let the time taken by a vehicle to pass over
piezoelectric roads be defined as:
= length of road laid with piezoelectric generator.
The intensity of energy generated due to mechanical
force of vehicles is calculated as:
 
Where U
= mechanical energy
λ = 0.078 and P represents the Power generated by
the respective vehicle class of mass M
and k =
Constant (to convert joules to kWh = 2.7778 × 10
kWh). Thus, P is the final output that the regression
model must predict.
2.2 Applications and Estimation of
Energy Savings
Figure 1: Different Scenarios where piezoelectric materials
can be implemented.
Piezoelectricity can be generated by placing the
piezoelectric tiles at various locations. The locations
must have either large amounts of vehicles or human
movements to harness energy in a profitable way.
Figure 1 depicts various scenarios where
piezoelectric tiles/plates can be installed. For
instance, a Dance club can meet 60% of its energy
requirement (A., 2019), because large numbers of
people are contained in small areas. The frequency of
the movements is very high making them one of the
ideal locations to place the piezoelectric plates. From
future perspectives, airports will be an ideal location
to lay piezoelectric plates.
This section will discuss the architecture proposed by
the authors, dataset used and results obtained by
GREECOPE: Green Computing with Piezoelectric Effect
performing tests.
3.1 Proposed Architecture
The paper proposes an architecture which will
estimate the amount of electricity that can be
generated by harnessing piezoelectric effect from
moving road vehicles. The process flow consists of
two parts: the AI model for prediction of energy and
the distribution of generated energy. The AI model
takes the output of the GREECOCO (Kshirsagar et
al., 2021) system as its input.
The GREECOCO system consists of the vehicle
detection and classification algorithm which is the
authors' previous paper. Traffic videos are fetched
from the installed cameras and vehicle detection is
performed using three algorithms: YOLOv3, Faster
RCNN and Mask RCNN. After detecting the vehicle,
it is recognized into five classes which are car, bus,
truck, motorcycle and bicycle. The vehicle count is
stored in a database for further analysis. We access
this database to get a count which will be an input to
our next module which predicts the energy generated.
Figure 2 shows the overall architecture of a
system. The output of the GREECOCO system which
is the individual as well as total count of four vehicle
classes (Bus, Truck, Car and Motorcycle) is
combined with features such as vehicle mass and the
speed of moving vehicles. After performing feature
extraction, mechanical energy is calculated. This
mechanical energy is given as an input to the
Artificial Intelligence (AI) module which predicts the
amount of energy generated by each vehicle class
during various time durations. SVM and GLM
approaches are used for energy prediction. The
prediction results are stored in a database for future
reference and energy analysis. The generated energy
will be distributed to charging stations for electric
cars and streetlights.
3.2 Dataset Details
The training dataset (Table 1) of the energy prediction
model was created using the NumPy library for 6
years from 2014 to 2020. The dataset consists of a
count of individual vehicle classes, as well as the total
Figure 2: Proposed Architecture for GREE-COCO.
SMARTGREENS 2021 - 10th International Conference on Smart Cities and Green ICT Systems
Figure 3: Dataset details for the training of SVM and GLM
count of vehicles from the GREECOCO system.
Also factors such as weight and speed of vehicle
are attributes of the dataset, along with force,
mechanical energy, and the expected energy
generated by each vehicle class. In the dataset, the
average mass of individual car, bus, truck and
motorcycle is 1000 kg, 7500 kg, 9500 kg and 170 kg
respectively. The data was used for the training of
SVM and GLM regression models, which predicts the
power generated for each vehicle class depending
upon the number of vehicles.
The testing dataset consisted of 10000 (416 days)
hours of data. Figure 3 demonstrates the cyclic nature
of each vehicle class’s count during an entire day.
This portrays the daily life situation of traffic
distribution throughout the day.
3.3 Data Pre-processing
Feature selection is a method of defining and
selecting a subset of input variables that the output
variable is most dependent. Our dataset consists of
40,000 sample, each with 15 input features, and we
employed Correlation Statistics (Sedgwick, 2012)
and Mutual Information Statistics (Beraha et al.,
2019) to perform selection.
The Correlation coefficient (r) is used to measure
the linear associativity between variables.
Here x
and y
are the values of sample dataset
and and are the mean values respectively. Here
the value r = 1 means a perfect positive correlation
and r = -1 means negative correlation.
Mutual information is a measure of dependence or
mutual dependence between two random variables.
The mutual information between two random
variables X and Y can be stated formally as follows:
Where I(X ; Y) is the mutual information for X and Y,
H(X) is the entropy for X and H(X | Y) is the
conditional entropy for X given Y.
Mutual information is always larger than or equal
to zero, where the larger the value, the greater the
relationship between the two variables. If the
calculated result is zero, then the variables are
Table 1: Dataset used for training SVM and GLM models.
GREECOPE: Green Computing with Piezoelectric Effect
Figure 4: Feature Selection Results by Correlation Statistics.
Figure 5: Feature Selection Results by Mutual Information Statistics.
3.4 Regression Models for Estimating
To predict the generated energy by the vehicle, AI
models using SVM and GLM have been designed.
Support Vector Machine (SVM) algorithm, is a
decision boundary where the distance between the
closest member of different classes is maximum.
Many nonlinear decision boundaries, and kernels are
modelled by the SVM algorithm. The major benefit
of using SVM is the robustness, especially in high-
dimensional space, against data overfitting. Another
common training technique for a diverse set of
regression models is Generalized Linear Models
(GLM). GLM enables to express the relation between
covariates X and response y in a linear, additive
manner. As a result, GLMs is a perfect option for real
world datasets which are not linear and
heteroscedastic and where we cannot predict the
model’s error distribution.
Figure 6: The expected energy by different class vehicles.
SMARTGREENS 2021 - 10th International Conference on Smart Cities and Green ICT Systems
Figure 6 illustrates the prediction of Car,
Motorcycle, Bus and Truck for 50 hours
respectively. The figures highlight the SVM and
GLM predicted energy vs. actual energy generated
for a particular set of vehicles during a specific hour.
Throughout the table, Car generated more energy
than the other vehicle classes. The energy generated
by a vehicle depends upon the amount of load it
carries. If a vehicle is not carrying any load, then the
amount of energy generated differs in a measurable
amount than the vehicle carrying enormous load.
Figure 7 the comparison of vehicle’s generated
energy in 2 scenarios with and without load is
illustrated. We can see there is a significant amount
of difference in two scenarios output.
Figure 7: Comparison of Voltage Generation of different
vehicles with load and without load.
3.5 Performance Analysis of Regression
Models on Varying Traffic Count
The table 2 below shows five different scenarios with
varying vehicle counts and estimation of energy
generated from our AI models.
Scenario 1 depicts a total vehicle count of 100
where the categorical count for each vehicle class is
further distributed as 45 for the vehicle class car, 10
for the vehicle class bus, 5 for trucks and 40 for
motorbikes. In scenario 1, hence the total energy
generated by each class would be, for car 0.477 kWh,
bus 0.796kWh, truck 0.504 kWh and motorbikes
0.072 kWh. The combined energy amounts to 1.84
kWh, which can be used to light up 1 LED bulb
consuming 100w for a total duration of 18.4 hrs.
Similarly, scenarios 2-5 depict energy generation
according to different traffic counts assumptions.
In scenario 5, same number of vehicles are
considered irrespective of the vehicle class, though it
is noticeable that larger difference in the generated
Table 2: Analysis of Generated Energy from varying
vehicle count.
Vehicle Count
Lighting Up of
LED Bulbs
Period (100 W)
100 (C:45, B:10,
T:5, M:40)
1.84 (CE:0.477,
BE:0.796, TE:
0.504, ME:0.072)
1 bulb for
500 (C:250,
B:100, T:50,
15.83 (CE:2.654,
BE:7.962, TE:
5.042, ME:0.180)
6 bulbs for
1000 (C:20,
B:150, T:440,
58.82 (CE:2.123,
24 bulbs for
5000 (C:1200,
B:900, T:550,
60 bulbs for
10000 (C:2500,
B:2500, T:2500,
482.26 (CE:26.541,
200 bulbs for
This part will explore the returns of the investment
and the challenges that the system may face in the real
4.1 Estimation of Return of Investment
We now illustrate the scenario of Motorway 50 (M50)
of Dublin, Ireland to demonstrate the returns obtained
by installing it with piezoelectric plates. The
construction cost of the entire motorway was €1
billion (Paschal Donohoe, 2020). Consider the
following requirements for infrastructure
development for the construction roads laid with
piezoelectric plates.
The estimated number of generators required for
a 1 km patch of a 2-way road are 13,333. Thus, the
entire 4-lane motorway will require 53,332
generators considering the cost of each individual
generator to be €25 (Garland, 2013). For a 4-lane road
covering a distance of 45.5 km we have the
approximate cost as 45.5×53332×25 which results in
€60,665,150 (approximately €60.6 million).
Moreover, labour and installation costs needed for
GREECOPE: Green Computing with Piezoelectric Effect
laying the piezoelectric tiles beneath the road will be
around € 5 million and €7 million respectively (Chew
et al., 2017), also an additional miscellaneous amount
of € 10 million is considered. Hence the entire budget
becomes €1 billion + €60.6 million + €5 million + €7
million + 10 million which amount to €1.082 billion.
The average power output obtained from the
piezoelectric generators can be estimated by
considering the following traffic scenario:
Average traffic volume is 420,000 per day
(Baker., 2020)
Average Speed of vehicles: 60kmph
Power generated 175176.4 ×10
kWh (per
vehicle) from formulas (1) - (5) discussed in
section 3.2
Total power generated from the average traffic
volume = 7357.4 kWh
To calculate power generated from the entire road
patch of 45.5 kms we have,
Total power generated = 45.5 × 7,357.4 × 365
= 122,188,020.5 kWh
Currently, the Government of Ireland charges 17.67
cent/kWh for the electricity, thus revenue generated
from the generated electricity would be
= 122,188,020.5 kWh × 17.67
= €21,590,623.22
= €21.6 million (approx.)
Hence, the amount invested to lay piezoelectric
plates on M50 road will be recouped in approximately
5-6 years with the added labour costs and other
requirements. Knowing that the average life of
piezoelectric tiles to be around 30 years strongly
implies that the income generated by adopting this
technology can have roughly 24 years of profits.
Figure 8: Total energy generated and energy consumed.
Now let's consider the scenario, for instance, on a
200m patch of a road with 100,000 vehicular traffic
each day. Let this stretch have around 9 street lamps
(High-Pressure Sodium (HPS) Lamp) installed. Now
assuming that these 9 lamps each utilize 225 W per
hour, the energy required to light up the 9 street lamps
for 2 months, assuming they are used for 10hrs per
day, is 1215kWh. If we lay piezoelectric tiles on this
200m patch, then the energy generated is 12,740 kWh
(2 months), which further can be used to light up the
lamps. The comparison of energy generated and
energy consumed for lighting 9 HPS lamps is
highlighted in figure 8. The unused generated energy
(11525kWh) can be stored in Energy storage devices.
In addition to this, it can be used for electric vehicle
charging stations or can be distributed through the
electrical grid.
4.2 Challenges
The energy harnessing system using piezoelectric
material produces green, safe, and renewable energy
which can be generated at a high scale. In-spite of all
these positive sides, the system needs to resolve the
issue with respect to storing the generated energy.
The batteries' efficiency to convert the energy is not
constant and sometimes it is low. The available
technology of the storage system (for instance
Batteries) highly influences the efficiency of the
energy harnessing system. It must be understood that
the rechargeable batteries have a fixed lifespan to
consider the option of recharging the depleted
batteries. In some cases, due to limited storage,
batteries can seldom store only 50% of the generated
energy. However, dependency upon batteries can be
reduced or eliminated by choosing an independent
method for the energy harvesting system on the roads.
Thus, the direct use of generated electric energy is
highlighted by the above approach Si.e., energy must
be used as it is produced. In such a self-sustaining
way, energy loss can be reduced. The other aspect to
address is around maintenance costs of the
piezoelectric plates or sensors which are embedded
within the pavement layers. Therefore, each time the
plates need to be repaired or changed, the asphalt
layers of the road must be removed and reopened.
Such repairs will incur additional costs. This issue can
be resolved by improving robustness of transducers.
This paper discussed the possibility of harnessing
piezoelectric energy from vehicular traffic in busy
roads. The authors design SVM and GLM regression
models to estimate the electricity that can be
generated from traffic counts of different road
vehicles. The authors comprehensively discuss the
potential return on investment by adopting this
SMARTGREENS 2021 - 10th International Conference on Smart Cities and Green ICT Systems
architecture for a sample express highway case. The
authors strongly validate the potential advantages that
can be incurred by adoption of this technology
through depiction of vast amounts of energy that can
be generated and potentially redistributed in various
settings. Finally, the authors through a comparative
tabulation with other popular renewable energy
sources bring forth the low maintenance and natural
environment preservation that is possible through
adoption of this technology.
This work is supported under World Bank Technical
Education Quality Improvement Programme Seed
Grant (Government. College of Engineering,
Aurangabad) and partially supported by Science
Foundation Ireland grant #16/IA/4605.
Reeker, G. 2014. Non-renewable resources and the limits
of economic Growth. GRIN Publishing.
Davies, L. 2017, December 21. Types and alternative
sources of renewable energy. EDF. https://www.ed
Borikar, J., 2017. Foot Step Power Generation.
International Journal for Research in Applied Science
and Engineering Technology, V(X), pp.1022-1023.
Kaur, H. 2017. An Insight Into Supervised Learning Using
Support Vector Machines. 2017. International Journal
of Recent Trends in Engineering and Research, 3(8),
McCullagh, P., &Nelder, J. A. (1989). Generalized Linear
Models (Chapman & Hall/CRC Monographs on
Statistics and Applied Probability) (2nd ed.). Chapman
and Hall/CRC.
Kour R, CharifA . 2016. Piezoelectric Roads: Energy
Harvesting Method Using Piezoelectric Technology.
Innov Ener Res 5: 132.
A. 2019. Science of Piezoelectricity: Generate Electricity
While You Groove on The Dance Floor. Science ABC.
Kshirsagar, M., More, T., Lahoti, R., Agaonkar, S., Jain, S.,
Ryan, C., &Kshirsagar, V. ,2021. GREE-COCO: Green
Artificial Intelligence Powered Cost Pricing Models for
Congestion Control. In Proceedings of the 13th
International Conference on Agents and Artificial
Intelligence - Volume 2: ICAART.
Sedgwick, P. 2012. Pearson’s correlation coefficient. BMJ
: British Medical Journal, 345.
Beraha, M., Metelli, A.M., Papini, M., Tirinzoni, A.,
&Restelli, M. 2019. Feature Selection via Mutual Infor-
mation: New Theoretical Insights. In 2019 International
Joint Conf. on Neural Networks (IJCNN), 1-9.
Paschal Donohoe. Retrieved December 10, 2020, from
Garland, R., 2013. Piezoelectric Roads in California.
Submitted as coursework for PH240, Stanford
Chew, Boon Cheong & Loo, Heoy & Bohari, Izyan &
Hamid, Syaiful & Sukri, Fatin & Kusumarwadani, Rini.
2017. Feasibility of piezoelectric tiles adoption: A case
study at Kuala Lumpur International Airport (KLIA)
Malaysia. AIP Conference Proceedings.
Baker, N., 2020, May 27. M50 busiest as traffic volume up
6% on national roads. Irish Examiner. https://www.
GREECOPE: Green Computing with Piezoelectric Effect