Content uploaded by Nigel Thomas
Author content
All content in this area was uploaded by Nigel Thomas on Dec 14, 2021
Content may be subject to copyright.
37th Annual UK
Performance Engineering
Workshop
UKPEW 2021
15th December 2021
Programme
9.30 Welcome and introductions
Session 1: 9.35 to 10.50 (Session Chair: Karim Djemame)
• Saad Alateef and Nigel Thomas, Energy consumption estimation for electric vehicles
using routing API data.
• Zhichao Ma (Fintrust Group), Han Wu (Fintrust Group), Aad van Moorsel, A Markov
Chain Model Developed To Investigate The Impact Caused By Open Banking
(Deposit Aggregators) in Financial Service Industry.
• Hassan Jari, Ali Alzahrani, Nigel Thomas, A Novel Indirect Trust Mechanism for
Addressing Blackhole Attacks in MANET
Break 15 minutes
Session 2: 11.05-11:55 (Session Chair: Nigel Thomas)
• Fatemeh Banaie Heravan and Karim Djemame, EDGENESS: Energy-efficient
Internet Architecture using a Serverless Platform
• Zhihao Dai (Warwick), Matthew Leeke (Warwick), Shuang-Hua Yang (SUSTech),
miniICS: A Lightweight Simulation Platform for Industrial Control Systems
12.00 Close
Energy consumption estimation for electric vehicles usning routing
API data
Saad Alateef1
School of Computing
Newcastle University
Newcastle upon Tyne, UK
Nigel Thomas2
School of Computing
Newcastle University
Newcastle upon Tyne, UK
Abstract
Electric vehicle (EV) range anxiety is an influential factor in electric vehicle’s low penetration into the transportation system.
There have been several developments on range estimation for electric vehicles, however, the studies which focus on determining the
remaining range based on the real-time publicly available data remain low. The majority of the current methods being employed
consider limited data collection and do not consider the most substantial factors that directly impact energy consumption. This
paper introduces a velocity model based on route information for the range estimation of electric vehicles. It uses publicly available
data sets obtained from several map services APIs and incorporates this data in the range estimation algorithm. Three map services
APIs were used to collect the data for an extended period, and then we analysed this data to extract the most representative data
to generate the velocity profiles. The paper uses MATLAB code and python libraries to process the representative data and apply
the velocity model. Moreover, we have integrated it into an electric vehicle model, including the battery, to estimate the power
demand for each trip and the remaining driving range. We observed that producing realistic driving cycles using public data is
possible; furthermore, it simulates the driving patterns and satisfies the constraints of the vehicle.
Keywords: Routing API, Electric vehicles, SOC estimation
1 Introduction
Electrifying transportation is one of the main targets for the transportation sector to reduce greenhouse emis-
sions in most countries [1]. However, Internal Combustion Engines (ICEs) are entirely dependent upon fossil
fuels and still the primary propulsion system in road transport globally. The increase in the dependency on
oil is considered significant as a result [2]. Therefore, there is an essential need to overcome this issue to
increase the sustainability of the transportation system and address the environmental issues. The demand
for electric vehicles has been increasing recently in the transportation markets, and it is expected to continue
to replace traditional vehicles in the next few decades. EVs are an intelligent solution for the planet and will
reduce gas emission significantly [3]. However, range anxiety is one of the main challenges that face electrifying
transportation, and it affects the adoption of electric vehicles.
In addition to the enormous advantage of reducing the levels of pollution EVs have, this invention has some
other benefits over conventional vehicles. These benefits include energy recovery when the battery restores
some of the energy due to braking, and the noise-freeness [4]. Regenerative braking is a crucial characteristic
1Email: s.alateef1@ncl.ac.uk
2Email: nigel.thomas@ncl.ac.uk
of EV when the generator returns the energy to the battery system due to braking. According to previous
studies, this feature is practical, especially in city driving and the daily commute. However, it is less effective in
motorway journeys, and long journeys [5]. Conventional vehicles consume more energy in city driving because
of the heat loss due to braking in contrast with EVs [3].
This paper aims to develop a velocity model using the publicly available routing data on specific routes.
It attempts to construct the speed profile for a specific journey between origin and destination using the
map API. After generating the potential realistic driving profile, we used a generic EV model to generate the
potential power demand for the trip. Hence we apply the state of charge estimation method to analyse the
impact of the route and traffic on the battery efficiency. This research concentrates essentially on developing
a data collection process using multiple maps service API. Many drivers rely on the GPS data provided by
map services to navigate to their destinations [6]. This paper uses the data collected from the drivers using
the map API. The first step of this paper involves exploring the routing information and using it to estimate
the energy consumption and improve the battery-powered vehicles’ efficiency. This research explores the data
of three different map information providers through their API. Google Maps API [7], HERE Maps API [8]
and TomTom Maps API [9] are the primary data sources in this research.
The amount of data collected from vehicles and drivers can significantly improve the range of electric vehicles
[10]. The battery management system (BMS) installed in electric vehicles senses the battery state of charge
and predicts the remaining range based on the battery status and some other data installed on the system
such as the vehicles’ specifications data. However, these data-sets do not consider the route information ahead.
Therefore, it uses the range values for its estimation. The proper use of the available data can improve the
driving range prediction and improve the energy consumption estimation.
In this paper we construct near to real-time velocity profiles to allow us to generate power profiles and estimate
the power consumption before performing the journey.
1.1 Energy consumption and driving cycles
Energy consumption in transportation systems has been a significant research and development topic recently
[11]. Previous work focused on how the driving behaviours affect the fuel consumption in internal combustion
engine (ICEs) vehicles [12]. In recent years, further studies have been conducted on the usage and consumption
of EVs [13]. These studies are characterised based on their methodology, and purpose [14]. In addition, some
researchers focus on the energy models of electric vehicles to improve the EV design [15], exploring the influential
factors on power consumption [16] and the influence of the driving patterns on the energy consumption and
the remaining driving range [17].
Whilst, there are many studies in the literature to improve the energy consumption of electric vehicles; there
is less research conducted on energy consumption based on the real-time velocity profile prediction. These
profiles are known as driving cycles for vehicles and generally defined as a series of points representing speed
versus time. The driving cycle is usually performed as a physical journey on a vehicle for various purposes and
based on various criteria [18].
Driving cycles developed in recent decades are used as a standard tool for estimating fuel consumption and
measuring the levels of air pollution produced by the transportation system. Many existing standard industrial
driving cycles such as NYCC, UDDS, and HWFET, have been used in some studies [19,20]. These driving
cycles are used as velocity profiles for validating the EV and battery models response. The current driving
cycles performed in unknown conditions and do not represent the real-time driving conditions. Some existing
studies developed methods to predict the driving profile [21,22], and each method relies on the nature of
the data used to develop this prediction method [23]. The map service API can help up to some extent to
develop and improve the real-time driving cycle construction methods. The API provides a wide range of route
information for any geographical location on the map and also considers the traffic situation. Even though the
API providers restrict developers from some features for commercial and competition reasons, it is still possible
to extract some valuable data to help to predict the journey and the velocity characteristics to improve the
range and energy estimation for electric vehicles. This approach makes it more convenient than performing
the physical journey considering many arrangements and set-ups such as a vehicle, driver and some equipment
making it a costly task [24].
2 Data collection process and analysis
2.1 Traffic data exploration
This section illustrates the process and the purpose of exploring the traffic data. In addition, the data collection
process and the challenges faced are also presented.
2
(i) Route Selection:
The main objective of collecting the data from the map service providers is to create a generic script that
gathers time-specific traffic data between two different Geo locations following a specific route. We have
specified the origin and destination on the map for two different routes that have different road structures.
These routes were sliced into multiple chunks so that we can collect more accurate data for each chunk.
Collecting the data for smaller segments is to separate the parts of the route that have possibilities of
speed reduction from more continuous high-speed such as motorways.
(ii) Data Analysis:
The data provided from the APIs are “duration”, “distance” and “segments”. Each segment profile
includes duration and distance. Since the distance and the time are known, we can calculate the average
speed for each segment and therefore, for the entire route. The plots for these raw collected data gives us
an idea of what the speed profile, as it presents the average speed for each segment of the route.
(iii) Data Manipulation:
Since the data obtained from the APIs are only average speed based on the duration in traffic and
distance of the segment, it provides a constant speed for each chunk of the road. Therefore, we introduce
some changes to those average speeds to reflect more realistic driving patterns. Therefore, it can represent
the velocity of the vehicle in each segment without altering the mean value of the speed provided from
the API data
2.2 Data collection
(i) Data collection methodology.
•Extracting the data from the API provider.
•Collecting data from the API response.
•Scheduling the collection process for specific times.
•Loading the data into a CSV format.
(ii) Source of traffic data.
•Google Maps API
The API products provided from Google Maps were used as follows:
·Distance Matrix API: This API allows us to get the travel distance and time for the entire route and
each identified segment. In addition, it allows us to obtain the estimated duration within the current
traffic.
·Directions API: Allows introducing the way-points which helps force the API to follow the route we
specify; it is also responsible for the mode of transportation, which is ”Car” in our case.
•TomTom Maps API
The API products provided from TomTom were used as follows:
·Traffic Flow API : This allows developers to request the travel time from the origin and destination
with respect to the real-time traffic.
·Maps API: This product gives an access to the API data every time we make a request.
·Routing API : This API gives highly detailed information about the route, with respect to directions
and travel mode.
•HERE Maps API
The API products provided from HERE Maps were used as follows:
·Routing API: This product informs the estimated arrival time between the origin and destination.
·Traffic API: This API is responsible for reporting the traffic flow, its consequences and the incidents
information.
·Way-points sequence API: This allows us to specify the way-points on the route to divide it to the
segments we require.
(iii) Extracting the time and speed data
•The data of the time taken during current traffic and the average speed calculated are added into
separate files for each journey. These files are formatted in two columns that show the time in seconds
for the whole journey versus the average speed at each second. These files are then processed to generate
possible velocity profiles.
In Table 1, the main features of the used map services API are illustrated.
3
Table 1
API features
The data was collected at multiple time-slots for each API. These slots were at 8:15am, 12:00pm, 16:45pm and
12:00am. This time selection was done to evaluate and analyse the differences between the peak traffic hours
and when it is quiet.
During each slot, the data is requested for an hour, and then loaded the data into CSV files in several rows.
The number of rows are dependent upon how many intermediate points were introduced. The data consists of
many columns starting from the date when the data was collected, until the average speed that was calculated
using the distance and the duration in traffic. Each row is a repetition of the same process during the specific
time we selected. Figure 1illustrates a step by-step-process of collecting the data through the APIs.
Fig. 1. Data Collection Process
3 Route based driving cycle construction
This section explains how the acceleration and deceleration is applied to the average speed data then add the
noise function to introduce some kind of variations to the speed profile wherever it is constant. In addition it
4
0
5
10
15
20
25
30
0500 1000 1500 2000 2500
Velocity (m/s)
Time (s)
API Data
Fig. 2. Mean velocity obtained from HERE Maps API
illustrates the method used to smooth the velocity curves.
3.1 Applying acceleration and deceleration between route segments
To smooth the transition of velocity between segments, we applied the acceleration and deceleration rate to
the beginning and ending intervals. Based on Nissan leaf’s 2019 [25] acceleration rate for 0-100 km/h, we
determine the maximum acceleration on the car. We consider that the acceleration and deceleration rates the
same.
Fig. 3. The initial driving cycle before the speed transition between segments
Using the data retrieved from the API, we obtain the initial driving cycle as shown in Figure 3. It is charac-
terised by sharp edges, corresponding with unrealistic significant speed changes. In addition it does not take
into account the technical constraints imposed by the vehicle and the road characteristics. Therefore, the final
driving cycle needs to be developed realistically before performing the energy consumption estimation.
The process of developing the driving cycle is implemented in iterative manner. In Figure 4, the driving cycle
shows three different segments which constant speeds. The velocity on the first segment is assumed to be at
speed V1, and since the recorded velocity on the second segment is higher than the vehicle’s velocity on the
second segment, the vehicle needs to accelerate gradually after exceeding point A. The determination of the
5
acceleration is based on the speed difference between V1and V2using the following equation:
a=(3.5, v2−v1≥10km
h
1
2(v2−v1), v2−v1<10km
h
(1)
After determining the acceleration, The time ∆tneeded for the vehicle to accelerate from the velocity in the
first segment V1to the following velocity V2can be calculated as:
∆t=v2−v1
a(2)
Fig. 4. The gradual acceleration added to the driving cycle
Calculating the distance ∆sthe vehicle needs during the accelerating process leads to the division of the
following segment into separated segments as shown in Figure 5
∆s=v1∆t+a∆t2
2(3)
The first segment has the length ds where the vehicle acceleration is applied until it reaches the speed V2.
The second segment has the length S2-ds when the vehicle’s velocity is constant and equals V2. The API
data speed data are often imperfect and inconsistent, it deviates from the real life conditions and constraints.
Therefore, the acceleration between velocities are not always feasible, in other words, for the above analysed
case of the acceleration from V1to V2, sometimes the distance that the vehicle needs to accelerate is longer
than the length of the following segment itself. To overcome this issue, the acceleration V2will not take place,
moreover, we reduce the speed on the following segment by small step ∆, and repeat the process where the
speed on the next segment is V2- ∆. This whole process is repeated until it satisfies the feasibility yielding
the final driving cycle as shown in Figure 5.
3.2 Adding noise function
To mimic a real driving cycle, we add noise to the intervals in which the speed is constant. The noise is
generated as uniformly distributed random numbers in the interval [a, b]. Considering that small variations in
speed are accepted, a and b are defined as functions of the maximum and minimum speeds of an interval i.
a=−5×1
minimum(vi)and b= 5 ×1
maximum(vi)(4)
6
Fig. 5. Final driving cycle after applying the acceleration method
The noise must not interfere in the travelled distance and the average speed in a plateau must remain unchanged.
Therefore, the mean of the noise must be zero. To ensure this condition, after the noise nis generated for N
samples, it is corrected as follows.
ni corrected =ni−n, i = 1,2,···, N (5)
3.3 Smoothing the sharp edges
As abrupt variations in speed remain after the acceleration method and noise adding, the last step consists of
smoothing the speed curve. We apply the LOESS (locally estimated scatter plot smoothing) method, using
4% of the samples for calculating smoothed values.
LOESS is a method of non-parametric regression that produces a smooth curve by locally fitting polynomial
functions. Thus, the fitted values are determined with neighboring subsets of data. LOESS, among other
methods, and the percentage of samples are chosen based on a qualitative evaluation of the final driving cycle
– the main criteria are the decrease of sharp edges, preservation of noise-induced variations and preservation
of the cycle when compared to its pre-processing shape. We determine that the cycle starts and ends at 0
m/s. To ensure a smooth transition, the speed curve is linearly interpolated from zero to the speed value of
an arbitrary point at the beginning and ending of the cycle.
After applying the previous methods, the represented driving cycles are generated as shown in the Figure
6. These figures present the velocity profiles for Google Map API after the representative driving cycles are
selected for the route. After we constructed driving cycles for each route and API, it is clear that the driving
cycle for each API is different at some points on the route and quiet similar at other points along the routes.
The generated driving cycles will be used in the next section to develop the power profile for the electric vehicle.
Hence, the energy estimation can be performed and the battery dynamics can be captured.
7
Fig. 6. Google Maps driving cycles for Route 1
4 Generating the power demand using electric vehicle’s dynamics
This section consider an electric vehicle model based on existing Nissan leaf to perform the power demand
generation and the state of charge estimation based on the data used on this research. With the vehicle speed
determined in the driving cycle, we calculate the power consumed to generate the vehicle, or, in case of braking,
the power provided back to the battery pack [25].
Battery discharge
(motoring)
Battery charge
(regenerative braking)
Battery Motor/
Generator
Transmission
system Wheels
Accessories
PbPin Pt
Pva
ηm,ηbrake
Ft
Pout
ηtr
Fig. 7. Electric vehicle power transition diagram
Ft(t) = Fr(t) + Fg(t) + Fd(t) + Fa(t) (6)
Starting at the wheels, the traction force Ftrequired for the vehicle’s motion is expressed by the sum opposing
forces, which is the rolling friction, grade resistance, aerodynamic drag, and acceleration force [26] and [25].
We consider the road slope α= 0 for the whole extension of the routes. Even though the road slop data is
available from some API map providers, it was not possible to obtain it accurately in this approach, since the
way-points were manually selected upon our previous knowledge of the routes, and this makes obtaining the
road slope information a complex task and inaccurate due to the uneven route segments length. In addition,
we implemented the rolling resistance and the force resisting the tires on the road surface.
8
4.1 Battery model dynamics and energy consumption estimation
The implementation of the battery model, was considering the Rint model proposed in [27]. This model
includes a voltage source Voc, representing the open-circuit voltage, in series with the parallel branch of internal
resistance. Any battery model can be implemented in this part of the research to estimate the state of charge
based on our power profiles. The current model is less complex and validated in previous studies such as in
[27].
+
_
_
+
V!"
𝑅!"#$%&
𝑅'()!"#$%&
𝐼V#!$%
Fig. 8. The equivalent circuit model based on Rint with two resistors in parallel
The current flow in the resisting branch is represented by ideal diodes. When the battery is discharging, the
diode in series with the discharging resistance (Rdischarge ) conducts the current; contrarily, in case of battery
charge, the diode conducting the current is in series with charging resistance (Rcharge). Given an initial state-
of-charge, we start by calculating open-circuit voltage Voc in terms of the SOC, where K,a,b,cand dare
constants.
Voc(t) = K−aSOC (t)−b1
SOC(t)
+cln(SOC(t)) = dln(1 −SOC(t))
The charging or discharging resistance Rsis a function of the SOC and is determined based on look-up tables
obtained from [25]. Then, the battery current is calculated by:
I(t) = Voc(t)−pVoc (t)2−4RsPb(t)
2Rs
(7)
The current is positive if the battery is discharging, and negative if it is charging. Finally, the SOC is estimated
with the coulomb counting method [28], in which the battery current is integrated over time to calculate the
transferred charge.
SOC(t) = S OC (t0)−1
CrZt
t0
)I∆τ(8)
SOC(t) = S OC (t−1) + I(t)∆τ
Cr
(9)
Where SOC(t) is the current state-of-charge, SOC(t0) is the initial state-of-charge, Cris the rated capacity,
Iis the current flowing in or out of the battery, t0is the initial time and t, the current time. Alternatively,
the SOC can be expressed in terms of its previously estimated value SOC (t−1) and the current for the time
interval of ∆τ= [t−1, t].
The equations of Voc,Iand SOC are applied iteratively over time to obtain the profiles for a full driving cycle.
5 Results
This section presents the power needed for some journey based on each API and route. It also shows the
battery voltage and the state of charge estimation.
9
Fig. 9. Lower bound for Google Maps data Route 2
Fig. 10. Lower bound for HERE Maps data Route 2
10
Fig. 11. Lower bound for TomTom Maps data Route 2
6 Conclusion and future work
noindent This paper constructed different driving cycles based on three API data and two different routes. A
data collection framework is developed which gathers the same data from different API and process the data
to generate realistic driving cycles. We divided the routes into slices using the route segmentation technique.
The data contain the distance of the journey, the time taken for the whole journey, the average speed for the
whole journey, and the waypoints. We developed a velocity model algorithm and introduced variations using
a random function based on Gaussian normal distribution.
After introducing some randomness to the mean data extracted from the APIs, we used the locally weighted
scatterplot smoothing function ”LOWESS” in MATLAB to fit a smooth curve to the randomised data and
eliminate any sharp edges. The data selection is based on data classification and statistical analysis. An electric
vehicle’s model based on Nissan leaf was implemented to calculate the power demand and the remaining range
for each cycle.
The results show that the driving cycles are within the range and the vehicle’s constraints are satisfied. More-
over, it simulates the driving patterns for each cycle. The results also show the variation between the different
data sources and the times for the data collection. The state of charge estimation for each cycle and route
varies for each route and data source. The route includes motorway driving, shows massive energy consumption
when the vehicle manages to drive at the highest speed limit and shows less energy consumption when the
traffic density restricts the speed. In contrast, the results also show less energy efficiency for city driving when
the traffic is dense because the journey time is longer.
The proposed velocity model can be implemented to any other data source with more flexibility for the route
segmentation. It can produce real-time velocity profiles construction without the need of collecting more data
for more extended periods. It was not possible to integrate weather API and Traffic lights detection due
to the restrictions in the map sources. However, this can be included when using open source API such as
OpenStreetMaps API, even though it has less accuracy. Further laboratory experiments will be conducted in
future to validate the results in this paper using Nissan leaf battery.
11
References
[1] John J Conti, Paul D Holtberg, Joseph A Beamon, A Michael Schaal, JC Ayoub, and James T Turnure. Annual energy
outlook 2014. US Energy Information Administration, page 2, 2014.
[2] Kojo Menyah and Yemane Wolde-Rufael. Co2 emissions, nuclear energy, renewable energy and economic growth in the us.
Energy Policy, 38(6):2911–2915, 2010.
[3] Xinkai Wu, David Freese, Alfredo Cabrera, and William A Kitch. Electric vehicles’ energy consumption measurement and
estimation. Transportation Research Part D: Transport and Environment, 34:52–67, 2015.
[4] Nina Juul and Peter Meibom. Road transport and power system scenarios for northern europe in 2030. Applied Energy,
92:573–582, 2012.
[5] Mike Knowles, Helen Scott, and David Baglee. The effect of driving style on electric vehicle performance, economy and
perception. International Journal of Electric and Hybrid Vehicles, 4(3):228–247, 2012.
[6] Lei Zhu and Jeffrey D Gonder. A driving cycle detection approach using map service api. Transportation Research Part C:
Emerging Technologies, 92:349–363, 2018.
[7] Google maps api (developers portal). https://cloud.google.com/maps- platform/. Accessed: 2019-09-03.
[8] Here Map. Here map apis. https://developer.here.com/products/maps/, 2019.
[9] TomTom. Tomtom map apis. https://developer.tomtom.com/products/maps- api/, 2019.
[10] Jun Bi, Yongxing Wang, Qiuyue Sai, and Cong Ding. Estimating remaining driving range of battery electric vehicles based
on real-world data: A case study of beijing, china. Energy, 169:833–843, 2019.
[11] Nikhil Kaza. Urban form and transportation energy consumption. Energy Policy, 136:111049, 2020.
[12] Abbas Fotouhi, Rubiyah Yusof, Rasoul Rahmani, Saad Mekhilef, and Neda Shateri. A review on the applications of driving
data and traffic information for vehicles energy conservation. Renewable and Sustainable Energy Reviews, 37:822–833, 2014.
[13] Xuewei Qi, Guoyuan Wu, Kanok Boriboonsomsin, and Matthew J Barth. Data-driven decomposition analysis and estimation
of link-level electric vehicle energy consumption under real-world traffic conditions. Transportation Research Part D: Transport
and Environment, 64:36–52, 2018.
[14] Cedric De Cauwer, Joeri Van Mierlo, and Thierry Coosemans. Energy consumption prediction for electric vehicles based on
real-world data. Energies, 8(8):8573–8593, 2015.
[15] Guido Wager, Mark P McHenry, Jonathan Whale, and Thomas Br¨aunl. Testing energy efficiency and driving range of electric
vehicles in relation to gear selection. Renewable Energy, 62:303–312, 2014.
[16] Enjian Yao, Zhiqiang Yang, Yuanyuan Song, and Ting Zuo. Comparison of electric vehicle’s energy consumption factors for
different road types. Discrete Dynamics in Nature and Society, 2013, 2013.
[17] Chris Bingham, Chris Walsh, and Steve Carroll. Impact of driving characteristics on electric vehicle energy consumption and
range. IET Intelligent Transport Systems, 6(1):29–35, 2012.
[18] Tim J Barlow, S Latham, IS McCrae, and PG Boulter. A reference book of driving cycles for use in the measurement of road
vehicle emissions. 2009.
[19] Hewu Wang, Xiaobin Zhang, and Minggao Ouyang. Energy consumption of electric vehicles based on real-world driving
patterns: A case study of beijing. Applied energy, 157:710–719, 2015.
[20] Saad Alateef and Nigel Thomas. Battery models investigation and evaluation using a power demand generated from driving
cycles. In Proceedings of the 12th EAI International Conference on Performance Evaluation Methodologies and Tools, pages
189–190, 2019.
[21] Emilia Silvas, Kobus Hereijgers, Huei Peng, Theo Hofman, and Maarten Steinbuch. Synthesis of realistic driving cycles
with high accuracy and computational speed, including slope information. IEEE Transactions on Vehicular Technology,
65(6):4118–4128, 2016.
[22] Ying Yang, Qing Zhang, Zhen Wang, Zeyu Chen, and Xue Cai. Markov chain-based approach of the driving cycle development
for electric vehicle application. Energy Procedia, 152:502–507, 2018.
[23] Peter Lipar, Irena Strnad, Martin ˇ
Cesnik, and Tomaˇz Maher. Development of urban driving cycle with gps data post
processing. Promet-Traffic&Transportation, 28(4):353–364, 2016.
[24] A Esteves-Booth, T Muneer, H Kirby, J Kubie, and J Hunter. The measurement of vehicular driving cycle within the city of
edinburgh. Transportation Research Part D: Transport and Environment, 6(3):209–220, 2001.
[25] Paolo Iora and Laura Tribioli. Effect of ambient temperature on electric vehicles’ energy consumption and range: Model
definition and sensitivity analysis based on nissan leaf data. World Electric Vehicle Journal, 10(1):2, 2019.
[26] Habiballah Rahimi-Eichi and Mo-Yuen Chow. Big-data framework for electric vehicle range estimation. In IECON 2014-40th
Annual Conference of the IEEE Industrial Electronics Society, pages 5628–5634. IEEE, 2014.
[27] Ricardo Maia, Marco Silva, Rui Ara´ujo, and Urbano Nunes. Electrical vehicle modeling: A fuzzy logic model for regenerative
braking. Expert systems with applications, 42(22):8504–8519, 2015.
[28] Martin Murnane and Adel Ghazel. A closer look at state of charge (soc) and state of health (soh) estimation techniques for
batteries. Analog devices, 2:426–436, 2017.
12
Title: A Markov Chain Model Developed to Investigate the Impact of Open Banking on the
Financial Service Industry.
Main Authors: Zhichao Ma (z.ma5@ncl.ac.uk)
Han Wu (han.wu@ncl.ac.uk)
Aad van Moorsel (aad.vanmoorsel@ncl.ac.uk)
Other Relative Authors: FinTrust Group, School of Computing, Newcastle University.
FinTrust Group Official Website: fintrustresearch.com
Abstract
Background: Open Banking allows regulated third parties to use financial data held by banks
to develop innovative products that benefit individuals and small businesses. The growing
popularity of Open Banking accelerates the competition amongst banks. Market instability can
therefore be caused by furious competition between banks and lack of control on a huge number
of FinTech applications. Banks should mitigate the potentially exposed risks (e.g., credit risk
and liquidity risk).
Research question: How to simulate the market movement where a customer’s account is
managed by a fintech platform (such as Deposit Aggregator)?
Methodology: A Markov Chain model was developed to simulate the entire process of market
movement. To comprehensively investigate the influence generated by fintech products and
bank’s response to the market turbulence, several banks were simulated in the model with
varied transition probability applied. Higher incoming transition probabilities were given when
there was a more competitive product offered. Simulation was conducted in Python to
determine the market movement over time.
Result: One of the main outcomes from this simulation work was that the number of customers
for each bank was reasonably predicted based on the transition probabilities of banks. This
simulation implied that transition probability was the key factor influencing the churn, given
the fact that fintech platform aims to seek the more profitable financial products.
Future works: Our longer-term aim is to determine whether the market is stable under Open
Banking. This requires additional detail to be represented in the states of the model. Vice versa,
if data is available, we can use the model and apply Hidden Markov Chain technique to
determine the parameters of the model and estimate long-term distribution of customers over
banks.
A Novel Indirect Trust Mechanism for Addressing Black Hole
Attacks in MANET
Hassan Jari, Ali Alzahrani, Nigel Thomas
(h.a.m.jari2@ncl.ac.uk, a.a.a.alzahrani2@ncl.ac.uk, nigel.thomas@ncl.ac.uk)
School of Computing, Newcastle University Newcastle-upon-Tyne, UK
ABSTRACT
Mobile Ad hoc Networks (MANETs) are emerging wireless network with many distinct
characteristics. MANET has potential to be useful in different commercial applications.
However, it is necessary to address various security issues related with MANET. Data routing
is one of fundamental operation in the MANET. This research work addresses security issues
related with MANET routing protocols. In this paper we consider the performance of
MANETs routing protocols subject to a blackhole attack. Trust management aims to protect
a network from malicious behaviour. This research work proposed a novel trust-based
routing for MANET, called ITAODV, which is derived from regular AODV protocol. The
proposed protocol uses indirect trust mechanism which takes into account the reliability of
each node for forwarding the packets. The performance evaluation of AODV and ITAODV
carried out using network simulator NS-3. The experimental results demonstrate
effectiveness of the ITAODV protocol against the black hole attack.
EDGENESS: Energy-efficient Internet Architecture using a
Serverless Platform
Fatemeh Banaie, Karim Djemame
School of Computing
University of Leeds
Leeds, UK
Abstract
Network management strategies are undergoing a transition from using the proprietary technology of a vendor toward the open-
source software modules with service automation. The first glimmer of this transformation is Software Defined Networking (SDN)
that increases the velocity of network evolutions by delivering new network capabilities. The monolithic design of current SDN
controllers aggregate all functions into a single and huge program that restricts its ability to deploy a new service, independent
of other services. To address the problem, this paper describes a modular, and micro-service based SDN architecture that applies
network programmability within the context of Network Function Virtualization (NFV) and explores how it could benefit from the
serverless computing paradigm. Serverless Computing could reduce cost by providing an adaptive and scalable service platform,
which is the main factor in energy efficiency.
Keywords: Network management, network evolutions, SDN, serverless platform.
1 Introduction
In recent years, a growing number of users and applications are starting to make use of the Internet to meet
the objectives regarding socio-economic development and quality of life. Hence, the current Internet has been
evolved into an immense commercial platform with unprecedented data traffic demands. This has led to the
ever increasing demands for security, social content distribution, mobility, and Quality of Service (QoS) that
are difficult to be met by the current Internet architecture. The service model of the current Internet is
host-to-host IP packet delivery, which is not best suited for today’s content and service-oriented usage [1].
Although several extensions have been added into the end-to-end service model of the Internet, however, it is
still extremely hard to support the increasing demands for performance, mobility, and multi-homing through
such host-to-host packet delivery [2].
Despite the Internet’s architectural stagnation, a significant technical development, Software Defined Net-
working (SDN), has emerged to accelerate the design and implementation of the next generation computer
networks [3]. SDN allows a remote software controller to program the forwarding states in a data plane owing
to a set of network policies to build flexible hardware and infrastructure management. Namely, SDN offers the
horizontal integration of the network by separating the control functions from those functions that perform the
packet processing mechanisms and details.
Realizing the promise of SDN requires intent-based management and control with guarantees on safety, scale,
and performance. However, the monolithic design of current SDN controllers aggregates all SDN controller and
its application into a single and huge program. This approach requires a specific set of programming interfaces
and services for each SDN controller. Hence, it restricts the development of the new applications/services by
making them dependent on a particular SDN controller and programing language.
1Email: K.djemame@leeds.ac.uk
Banaie, and Djemame
Control plane
Data plane
Application plane
Network App Network App
Network Operating System
(NOS)
Infrastructure Layer
Northbound API
Southbound API
. . .
Fig. 1. The basic architecture of SDN.
To address the mentioned limitations, in this paper, a modular and micro-service architecture is designed to
divide the controller functionalities and applications into a set of cooperative micro-services. Thus, it provides
an easy way to add new services and features to the SDN controller. Aside from the quick adaptability of new
services, each of these micro-services can be run in a container. The container-based SDN platform enables the
orchestration tasks for modules, such as automation of service delivery, scaling up/down, and fault isolation.
Edge computing [4] allows Internet’s services to be implemented at network edges, known as service nodes
(SNs) [1]. The service nodes usually involve simple processing such as packet/request processing/handling,
which are short-live and event-driven. Thus, network services have the potential of being executed on a
serverless platform for deploying adaptive and scalable services with varying demands. Serverless computing
decreases operational costs by offering a finer granularity of resource provisioning that is a driving factor for
energy efficiency.
Before continuing, we review the reasons why we need to revisit Internet architecture. The rest of the paper
is organized as follows: the next section summarizes the current Internet shortcomings. Section III explains
the concept of modular SDN in the serverless platform. Section IV considers future research directions.
2 Why Revisit the Architectural Strategies?
The Internet represents one of the most successful projects in developing information infrastructure. However,
the current Internet limitations are based on its design principles and host-to-host communication model [5].
One challenging aspect of Internet is that the traditional IP networks are complex and difficult to manage
due to the vertically integrated network devices [6]. Specifically, network devices consist of controlling (which
specify the way to manage the incoming data) and data forwarding functionalities (which specify the packet
processing strategies). In vertically integrated network equipment, all functions are integrated into a single
device, which does not provide flexibility to the network manager. SDN improves network capability and
flexibility by separating controlling functions (such as routing and configuration management) from underlying
hardware (such as switches, routers, etc.). With a logically centralized network view, developers can move
from pairwise protocols to distributed algorithms with guaranteed flexibility and performance.
Moreover, today’s usage has changed from host-to-host packet delivery to content distribution, multi-
homing, and cloud services. Although the simple best-effort service model of the Internet meets the require-
ments of particular applications such as email and web access, It is not well suited to today’s applications such
as audio/video streaming that demands low latency. To this end, SDN can be used to solve the issues of QoS
guarantee in Internet [7], by proving on-demand services for guranteed QoS. These additional services can be
deployed and managed at Internet edges, thus, the Internet core can be only used for packet delivery [1]. This
approach provides flexibility and QoS guarantee support of Internet for new applications, and allows for many
optimization from fast deployment and resource management point of view.
3 A FaaS Architecture for Modular SDN
In this section, we describe the modular, and micro-service based SDN architecture that applies network
programmability within the context of Network Function Virtualization (NFV), and explore how it could
benefit from the serverless computing paradigm [8]. We consider a modular SDN architecture deployed on a
serverless platform as illustrated in Figure 2. For scalability and fault isolation, we propose to disaggregate
the SDN functionalities into a set of cooperating micro-services. Consequently, SDN core services provide
minimum required functionality, and the other services can be provided by external applications.
2
Banaie, and Djemame
SDN core functions
Topology Service Flow Service
Inventory Service
Statistics Service
SB API (e.g. OpenFlow)
NB API
Firewall Load
balancing Intrusion
detection
. . .
SDN Applications
Switch #2 Switch #n
. . .
OpenFlow Swithes
Switch #1
Load
balancing
Energy
consumption
analysis
Functions
placement
Kubernetes
Node #1
Node #2
Node #n
. . .
Control plane
Data plane
Serverless platform
Fig. 2. FaaS framework for modular SDN.
SDN promises to decouple the control plane and data plane for scalability and easier network management.
The control plane consists of SDN core functionalities (e.g. topology service, flow service, inventory service,
etc.), and management applications (e.g. firewall, load balancing, routing, monitoring, etc.), which communi-
cate requirements via Northbound API (NBI). The data plane consists of forwarding elements (i.e. switches
and routers) and uses OpenFlow [9] as Southbound API (SBI). SDN controller operates by serving events
(which are defined as any change in the network) from both the southbound and northbound Application
Program Interfaces (APIs).
As the SDN controller is event-driven and modular, it can be deployed on a serverless platform. These
modules can be implemented as a set of independent functions (network functions) that leverage the benefits
of the serverless computing paradigm in providing on-demand scalability and efficient resource management.
Serverless computing, also referred to as Function as a Service (FaaS), provides a platform to develop appli-
cations as a set of independent functions. It accelerates application deployment by eliminating the need for
managing infrastructure, thus, can be well suited for NFV architecture. Namely, FaaS eliminates the man-
agement burden of resource allocation (i.e. choosing the right time as well as the right type of VMs, and
containers) and further reduces costs. Therefore, users only provide network functions (which are SDN micro-
services) to this platform. Upon receiving an event from Application Program Interfaces (APIs), the platform
start executing the services automatically by deploying new instances. Network developers no longer need
to consider function deployment, management, and scaling issues. Besides reducing the management burden,
network service providers are charged according to the number of events in the NFV context. Accordingly,
they only pay for what they use at a very fine granularity [10].
The reduction in the execution time and the average resource usage of these micro-services, allows for many
optimizations from the resource management point of view. Since energy consumption has become a vital issue
lately, employing edge computing with low-powered computing resources contributes to a more energy-efficient
system development. In the context of a serverless computing, a microservice is essentially a proxy for energy
usage as a unit of (serverless) compute, making functions instantiation and orchestration significantly energy
and resource efficient [11].
4 Conclusion and Future Research Direction
Software Defined Networking (SDN) has changed network control and management strategies from expensive,
proprietary hardware to open-source, software-centric implementation of middleboxes. In this regards, the
monolithic architecture for SDN controllers integrates all SDN functionalities into a single, gigantic program.
However, as discussed, a good system design needs modularity, so, we propose a modular, micro-service SDN
architecture to tackle the limitations of monolithic architecture. The modular architecture disaggregates con-
troller software into a minimum viable controller core cooperated with a set of independent micro-services. We
also explore the suitability of deploying a micro-service approach on a serverless platform. Serverless comput-
ing decreases cost and energy consumption, as it provides on-demand and efficient resource management. To
achieve our goal in migrating from a monolithic approach to a modular, micr-oservice architecture, a mech-
anism must be devised to initiate service instances upon event arrival from APIs. In future work, we are
3
Banaie, and Djemame
planning to implement and evaluate our solution on a widely used serverless framework.
Acknowledgement
The authors would like to thank the European Next Generation Internet Program for Open INTErnet Reno-
vation (NGI-Pointer 2) for supporting this work under contract 871528 (EDGENESS Project).
References
[1] Balakrishnan, H. and et.al, “Revitalizing the Public Internet By Making it Extensible”, ACM SIGCOMM Computer
Communication Review, 51(2), 2021.
[2] McCauley, J., Harchol, Y., Panda, A., Raghavan, B., and Shenker, S., “Enabling a Permanent Revolution in Internet
Architecture”, In ACM SIGCOMM, 2019.
[3] Jain, V., Yatri, V., Kapoor, K., “Enabling a Permanent Revolution in Internet Architecture”, Journal of High Speed Networks,
25, pp.1-40, 2019.
[4] Srirama, S. N., Buyya, R., “Fog and Edge Computing: Principles and Paradigms”, John Wiley & Sons, 2018.
[5] Victor-Ikoh, M.I., Kabari, G., “Internet Architecture: Current Limitations Leading Towards Future Internet Architecture”,
IJCSMC, 10(5), pp.102-112, 2021.
[6] Kreutz, D., Ramos, F. M. V., Verssimo, P. E., Rothenberg, C. E., Azodolmolky, S. and Uhlig. S., “Software-defined networking:
A comprehensive survey”, Proceedings of the IEEE, 103(1), pp. 14-76, 2015.
[7] Hu, C., Wang, Q., Dai, X., “SDN over IP: Enabling Internet to Provide Better QoS Guarantee”, International Conference on
Frontier of Computer Science and Technology, 2015.
[8] Aditya, P., and et. al, “Will Serverless Computing Revolutionize NFV?”, Proceedings of the IEEE, 107(4), pp. 667-678, 2019.
[9] Singhvi, A., Khalid, J., Akella, A., Banerjee, S., “SNF: Serverless Network Functions”, ACM Symposium on Cloud Computing,
2020.
[10] Shen, j., Yu, H., Zheng, Z., Sun, C., Xu, M., Wang, J., “Serpens: A High-Performance Serverless Platform for NFV”,
IEEE/ACM 28th International Symposium on Quality of Service (IWQoS), 2020.
[11] Djemame, K., “Energy Efficiency in Edge Environments: a Serverless Computing Approach”, Proceedings of the 18th
International Conference on the Economics of Grids, Clouds, Systems and Services (GECON), Lecture Notes in Computer
Science 13072, Springer, September 2021.
4
A Lightweight Simulation Platform for Industrial
Control Systems
Zhihao Dai1, Matthew Leeke1, Shuang-Hua Yang2
1University of Warwick, Coventry, United Kingdom
{zhihao.dai, matthew.leeke}@warwick.ac.uk
2Shuang-Hua Yang, Southern University of Science and Technology, Shenzhen, China
yangsh@sustech.edu.cn
Abstract: It has been more than a decade since Stuxnet ushered in a new wave
of cyberattacks targeting Industrial Control Systems (ICSs) and, as a conse-
quence, an increased focus on ICS security. Simulators are a powerful tool when
studying offense-defense dynamics, obviating the risk and perils of interacting
with a real-world process when developing security mechanisms. However, open
source ICS simulators remain scarce, with available solutions falling short in
their modelling. In this work we present a lightweight framework for simulating
ICSs. The framework is component-wise, self-contained, and network-adaptive
via a Controller Area Network bus to facilitate inter-device communication. We
verify the simulation of process and control under three scenarios established in
the wider literature. Finally, we demonstrate the applicability of the framework
in modelling a variety of attacks. Being open source, component-centric, and
self-contained, the framework allows practitioners to simulate control systems,
replicate perceived threats, and, thereby, adapt to new challenges with reduced
risk, cost, and delay.