Development of a driving cycle to evaluate the energy economy of electric vehicles in urban areas

Applied Energy 177 (2016) 165–178

Contents lists available at ScienceDirect

Applied Energy journal homepage: www.elsevier.com/locate/apenergy

Development of a driving cycle to evaluate the energy economy of electric vehicles in urban areas John Brady, Margaret O’Mahony ⇑ Centre for Transport Research, Department of Civil, Structural & Environmental Engineering, Trinity College Dublin, The University of Dublin, Dublin, Ireland

h i g h l i g h t s
Development of a driving cycle to evaluate energy economy of electric vehicles.
Improves on existing driving cycles by using real world data from electric vehicles.
Driving data from different road types and traffic conditions included.

a r t i c l e

i n f o

Article history: Received 3 February 2016 Received in revised form 12 May 2016 Accepted 14 May 2016

Keywords: Electric vehicle Driving cycle Battery range

a b s t r a c t Understanding real-world driving conditions in the form of driving cycles is instrumental in the design of efficient powertrains and energy storage systems for electric vehicles. In addition, driving cycles serve as a standardised measurement procedure for the certification of a vehicle’s fuel economy and driving range. They also facilitate the evaluation of the economic and lifecycle costs of emerging vehicular technologies. However, discrepancies between existing driving cycles and real-world driving conditions exist due to a number of factors such as insufficient data, inadequate driving cycle development methodologies and methods to assess the representativeness of developed driving cycles. The novel aspect of the work presented here is the use of real-world data from electric vehicles, over a six month period, to derive a driving cycle appropriate for their assessment. A stochastic and statistical methodology is used to develop and assess the representativeness of the driving cycle against a separate set of real world electric vehicle driving data and the developed cycle performs well in that comparison. Although direct comparisons with internal combustion engine driving cycles are not that informative or relevant due to the marked differences between how they and electric vehicles operate, some discussion around how the developed electric vehicle cycle relates to them is also included. Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction and background Using electricity for vehicle propulsion offers the possibility to substitute oil with a secondary energy source. This could ensure the security of energy supply and a broad use of renewable and carbon-free energy sources in the transport sector which could assist global CO2 emission reduction targets. Electric vehicles (EV) produce less effective CO2 per kilometre (i.e. including CO2 emitted from electricity generation) travelled and produce no local pollution such as PM10 and NO2 [1,2]. Recent research on electric vehicles is broad ranging. Onat et al. [3] studied vehicle options across 50 US states taking into account

⇑ Corresponding author. E-mail addresses: [email protected] (J. Brady), [email protected] (M. O’Mahony). http://dx.doi.org/10.1016/j.apenergy.2016.05.094 0306-2619/Ó 2016 Elsevier Ltd. All rights reserved.

state specific average and marginal electricity generation mixes, regional driving patterns, and vehicle and battery manufacturing impacts. In addition they evaluated the widespread use of solar energy to charge EVs and plug-in hybrids (PHEV). EVs were found to be the least carbon intensive vehicle option in 24 states while hybrid EVs were found to be the most energy-efficient option in 45 states. Mallouh et al. [4] developed a model to compare, using experimental data that was recorded from a testing vehicle (taxi) running in the streets of Amman city, an ICEV with a hybrid fuel cell/battery vehicle by replacing only the powertrain and keeping all other parts the same. Their simulation results confirmed that hybrid FC/battery vehicles have superior performance in terms of fuel economy, drivability, emissions, and efficiency, when compared with ICEVs. Saxena et al. [5] show that the energy storage limits of today’s EVs are outweighed by their high efficiency and the fact that driving in the US seldom exceeds 100 km of daily travel. The normal

166

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

daily travel of 85–89% of drivers can be satisfied with EVs charging with standard 120 V wall outlets at their home only. 77–79% of drivers on their normal daily driving will have >60 km of buffer range remaining for unexpected trips. Similar findings were noted by Weldon et al. [6]. Wikstron et al. [7] presents findings from a 3 year study of 550 EVs and their users in Sweden. They found that winter conditions seem to result in an unjustified decrease in use and a substantial share of battery capacity is redundant. They found that this was not due to the technical constraints of the vehicles but concerns of the drivers using the EVs in those conditions. Morrissey et al. [8] showed that the charging behaviours of EV users vary depending on the location of the charging infrastructure. Weldon et al. [9] showed that the environmental impacts of EVs in Ireland are highly influenced by the charging behaviours of individual users, and night-time charging was found to produce the largest environmental impact as a result of grid management decisions. Meng et al. [10] found that frequency instability caused by intermittent wind generation is reduced by the frequency response from the EV clusters. Large scales of EVs utilized as a demand response resource can promote the development of wind generation in the Great Britain power system. Schill and Gerbaulet [11] examine the impact of future scenarios of EVs on the German power system. They found that the impact on the load duration curve strongly differs between charging modes. They also found that the overall energy requirements of EVs should not be of concern to policy makers for the time being whereas their impact on peak loads should be. They also suggested that policy makers should be aware that cost optimised charging not only increases the utilization of renewable energy but also of low cost emission intensive plants. Saxena et al. [12] used powertrain modelling to estimate that average city energy use is 33 W h/km for electric scooters, 84 W h/km for low power 4-wheel electric vehicles and 123 W h/ km for high power electric 4-wheeler vehicles. Seedam et al. [13] developed an onboard system for installation on a motorcycle to measure the on-road driving pattern. The developed onboard system was applied to collect the on-road driving pattern of the motorcycle driving along the road network of the Khon Kaen city, Thailand for developing a motorcycle driving cycle. Rangaraju et al. [14] used real-world energy consumption data for an environmental assessment of electric vehicles compared with diesel and petrol vehicles. The influence of charging profile on the well to tank emissions of EVs is discussed by using hourly emissions and different possible peak and off-peak charging time frames. The study noted the importance of taking the driving behaviour of users and auxiliary energy consumption into account. In the absence of an electric vehicle driving cycle they used the New European Driving Cycle for the assessment. The results revealed that the auxiliary energy consumption is responsible for nearly 1/3 of the well to tank emissions. Wang et al. [15] found that electric vehicles in Beijing, including HEVs, PHEVs and BEVs, yield more fuel reduction benefits than in the U.S. because of the severe driving conditions and short driving ranges. They also confirmed that the Chinese current suggested label values based on NEDC cycle underestimate the fuel consumption of vehicles and fuel reduction benefits of electric vehicles in Beijing. They point to the importance of developing and using real-world driving cycles in designing and evaluating electric vehicles; a gap the research presented here addresses. The design of efficient powertrains and energy storage management systems for EVs relies on an in-depth understanding of realworld driving conditions. Driving cycles have been developed to provide velocity–time profiles that are intended to be representative of real-world driving conditions. They are then used to assimilate driving conditions on a laboratory chassis dynamometer or in a vehicle simulation model. The battery capacity, battery chemistry

and the sizing of electrical components in the drivetrain are all dependent on the desired driving range of the vehicle. The peak power demands of a cycle influence the size of the battery whereas battery state of charge fluctuations influence battery health and thermal management [16]. In addition to playing an important role in design, driving cycles also serve as a standardised measurement procedure for the certification and evaluation of the fuel economy, emissions and driving range of emerging vehicular technologies. Furthermore, real-world driving cycles are required for realistic lifecycle analyses and for evaluating the impacts of EVs on the electricity grid. Existing driving cycles have been designed such that they can be applied to a variety of vehicles irrespective of the intended real-world operating conditions of the vehicle. There are two types of driving cycles, transient cycles such as the Federal Test Procedure (FTP-75) [17] and modal cycles such as the New European Driving Cycle (NEDC) [18]. The primary difference is that modal cycles are a compilation of constant acceleration and constant velocity periods, whereas transient cycles involve many velocity variations, typical of on-road driving conditions. There are two categories of driving cycles, legislative and non-legislative. Legislative driving cycles such as the NEDC and the FTP-75 are used by regulatory authorities to certify a vehicle’s emissions and fuel economy within their respective jurisdictions. Non-legislative driving cycles such as the Hong Kong cycle [19], the Edinburgh cycle [20], the Athens cycle [21], the Toronto waterfront cycle [22] and the Singapore cycle [23] have broad applications in research from vehicle design to life cycle analyses. Driving cycles are useful for comparison purposes because they provide an estimation of fuel economy, emissions and driving range. However, there is poor correlation with real-world driving conditions and their effects on fuel consumption and emissions, particularly in relation to modal cycles. There are significant variations in real-world driving conditions compared to test procedures and this variation causes a significant difference in emissions, fuel economy and driving range in real-world operating conditions [24]. Tzirakis et al. [21] developed a driving cycle for Athens using data collected from an internal combustion engine vehicle (ICEV). Depending on the vehicle tested, fuel consumption and emissions were found to be 9–79% and up to 300% higher respectively than those observed over the NEDC. Seers et al. [25] developed two driving cycles for utility vehicles using data from on-board loggers, and revealed a major difference of 31% in fuel consumption over the FTP-75 cycles. The German Ministry of Transport, Building and Urban Development measured the fuel consumption of more than one hundred cars and found that the majority of the vehicles consumed 25% more fuel and thus emitted more CO2 emissions than certified [26]. It was reported that 40% of the cars exceeded their certified limit, while 2% of the vehicles had a fuel consumption of up to 70% higher than certified. Real-world driving conditions in the US have also been analysed. Fellah et al. [27] analysed 110 real-world cycles in Kansas City and found that real-world cycles are more aggressive than the American certification cycle, the Urban Dynamometer Driving Schedule (UDDS), resulting in a larger energy requirement per unit distance travelled. Tate et al. [28] used 621 GPS driving cycles from Southern California to assess the performance of plug-in hybrid electric vehicles (PHEV). The authors found that the associated power and speed values of the driving samples were higher than those associated with the UDDS cycle. It was noted that 94% of vehicles have a larger average energy consumption per unit distance travelled in real-world driving conditions compared to the UDDS and the Highway Fuel Economy Test cycle (HWFET) [29]. Patil et al. [30] simulated a PHEV over real-world GPS driving cycles logged in south-eastern Michigan. It was found that 90% of the trips in the dataset consumed more fuel per mile than the

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

UDDS and HWFET cycles. Kang and Min [31] conducted a dynamic simulation of a fuel cell hybrid vehicle using the FTP-75 driving cycle. Gonder et al. [32] used GPS driving cycle data collected from a fleet of 227 vehicles in the State of Missouri to simulate the fuel and electricity consumption of PHEVs. The results were compared against the HWFET and the FTP-75 cycles. The modelled fuel consumption was lower for real-world driving cycles than the certification cycles whereas the electricity consumption was higher for the real-world cycles than the certification cycles. The author noted that these findings suggest that the driving cycles do not capture the range of speeds and accelerations typical of realworld driving and that these differences can significantly affect the energy economy of PHEVs. Shahidinejad et al. [33] state that there are many concerns about the problems inherent in existing driving cycles such as the underestimation of cruise, acceleration and stop-and-go activities in different velocity brackets. They concluded that existing cycles do not completely emulate the realworld power demands of vehicles. There are various PHEV powertrain configurations such as serial, parallel and serial–parallel. Karabasoglu and Michalek [34] detail the various PHEV powertrain configurations. In the serial configuration an engine turns a generator which generates electricity to drive an electric motor to turn the wheels; the parallel configuration is capable of transmitting torque to the wheels from two different energy sources; and the serial-parallel uses a planetary gear device to operate both in series and parallel. Charge sustaining HEVs have two complementary drive systems, a combustion engine and an electric motor with a battery. Both the engine and the electric motor can drive the transmission at the same time. HEVs cannot be recharged from the electricity grid, their energy comes from gasoline and from regenerative braking. A vehicle’s power management system blends the power from the electric motor and the internal combustion engine (ICE) to provide propulsion energy. Each type of powertrain configuration requires a substantially different power blending optimisation technique. Rask and Rousseau [35] demonstrated that the energy economy of a blended PHEV is particularly sensitive to driving cycles, therefore the optimum battery management system is best determined by design using realistic driving cycles [36]. Kwon et al. [37] analysed the impact of driving cycle aggressiveness on the energy consumption of a vehicle using a simulation of a midsize parallel hybrid PHEV. The author evaluated the outcome of sizing the electric motor and battery to follow the urban dynamometer driving schedule (UDDS) cycle and six additional standard driving cycles. The results indicated that a PHEV’s design is directly influenced by the choice of driving cycle. Certification cycles are intuitive choices for vehicle design because they are the driving cycles that are used by legislative authorities for emissions and fuel economy certification. However, designing vehicles to a fixed driving cycle results in a suboptimal vehicle for any driving environment that varies significantly from the driving cycles used in the design. Furthermore, Dubarry et al. [38] suggest that the lack of comprehensive driving cycles to allow for the benchmarking of vehicle and battery performance could undermine the development of EVs and this could also have significant impacts on the calculated life of the battery. There are generally two approaches used to develop driving cycles. In the first approach model driving cycles are developed from a number of constant acceleration and constant velocity phases. Aggressive driving cycles result in higher energy consumptions than non-aggressive driving cycles and thus impact on the driving range of an EV. Therefore, the model approach to developing driving cycles is not suitable for developing driving cycles representative of the real-world driving conditions due to the smooth acceleration and deceleration characteristics of the cycle [39,40]. The second approach uses collected on-road driving data to

167

develop transient driving cycles and is used in the research presented here. Discrepancies between existing driving cycles and real-world driving conditions generally exist due to insufficient data, inadequate driving cycle development methodologies and methods to assess the representativeness of developed driving cycles of real-world driving conditions. Wang et al. [15] pointed to the importance of developing and using real-world driving cycles in designing and evaluating electric vehicles and this paper addresses this need specifically. Rangaraju et al. [14] mentions the absence of an appropriate electric vehicle driving cycle. The novel aspect of the work presented here is the use of real-world data from electric vehicles, over a six month period, to derive a driving cycle appropriate for electric vehicle evaluation. The work presented here uses a large dataset collected from EVs operating in the real-world and a proven driving cycle synthesis methodology to develop and assess the developed electric vehicle driving cycle. The cycle presented in this paper is designed for the Greater Dublin Area (Ireland); a medium sized European city. The paper is organised as follows: in Section 2, the data collection and processing is explained. In Section 3 the driving cycle development is presented and assessed. The conclusions are summarised in Section 4. 2. Methods The field data used in this study were collected as part of a nationwide EV demonstration project in Ireland. Seven Mitsubishi iMiEV battery electric vehicles (BEV) were trialled in households of employees of the Electric Supply Board (ESB), an Irish electricity utility company, in the Greater Dublin Area for a six month period. The participants were selected from different income brackets, education levels, genders and geographical areas in order to achieve a representative sample of the drivers in the area. The data loggers installed in the vehicles were configured to read information from vehicle sensors available on the vehicle’s CAN (Control Area Network) bus and to store these data in the logger’s internal memory along with the vehicle’s GPS position. GPS data and CAN bus messages were logged every five seconds and every one second respectively when the vehicle ignition was on. Specifically the vehicle’s velocity was logged every second from the CAN bus. The data of a total of 1485 journeys are used in this study. The geographical areas in Dublin in which the data were logged are shown in Fig. 1. 2.1. Data processing The instantaneous acceleration was computed directly from the measured velocity data using numerical differentiation. However, in practice, the velocity data was found to contain intermitting errors. These errors could lead to the computed instantaneous acceleration falling outside the feasible capabilities of the vehicle (Fig. 2(a)). Rakha et al. [41] investigated the suitability of a number of smoothing techniques on GPS velocity data and found that the Epanechnikov density kernel smoothing algorithm performed well compared to the other investigated methods. Hellinga [42] also applied the Epanechnikov density kernel smoothing algorithm to GPS velocity data in the development of a fuel consumption and emissions model for internal combustion engine vehicles (ICEVs). In this study, the Epanechnikov Kernel smoothing algorithm, with a bandwidth of three, was applied to the logged velocity data and the instantaneous acceleration rates were computed on the basis of the smoothed velocities. Fig. 2(b) illustrates the application of the smoothing algorithm to a segment of a cycle. However, the smoothing algorithm did not always guarantee that the acceleration

168

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

Fig. 1. Map of the data collection region.

Fig. 2. (Left) Acceleration rate computed from raw velocity data. (Right) Raw velocity and Epanechnikov kernel smoothed velocity data.

performance capabilities of the vehicle were not exceeded. Riches [43] reports the acceleration performance capabilities of the Mitsubishi iMiev vehicle in various velocity intervals. Subsequent to the application of the smoothing algorithm, the velocity and acceleration measurements of each cycle were inspected and the cycle was excluded from the analysis if the performance capabilities of the vehicle were exceeded. Following this, the energy economy of each driving cycle was simulated using Autonomie [44], a vehicle simulation package developed by the Argonne National Laboratory (ANL). The Autono-

mie software has been validated for several powertrain configurations and vehicle classes using vehicle test data from the US Department of Energy Vehicle Technologies Argonne’s Advanced Powertrain Research Facility (APRF) [45–48]. A default BEV powertrain which was available pre-programmed into Autonomie was modified to replicate the drivetrain of a Nissan Leaf vehicle using published technical specifications. The specifications of the modelled vehicle are available in Appendix A. (The efficiency of 95% referred to in the appendix is for the motor). The model was validated by comparing the energy economy computed

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

by the software to the certified energy economy of the Nissan Leaf. The energy economy is defined as the total energy consumed divided by the distance travelled (W h/km). The published driving range and energy economy of the Nissan Leaf as stipulated by Regulation UN ECE 101 are 175 km and 173 W h/km respectively [49]. This implies that it requires approximately 30.3 kW h (175 km  0.173 kW h/km) to fully charge the battery. The battery capacity of the Nissan Leaf is 24 kW h, which implies that the actual charging efficiency is approximately 80%. Ignoring the energy losses due to charging, the energy economy of the Nissan Leaf is approximately 137 W h/km (24 kW h/175 km). The simulated energy economy of the modelled powertrain over the NEDC in Autonomie is 144 W h/km, a 5% deviation from the certified energy economy. The modelled powertrain was simulated over all the recorded driving cycles to simulate the energy economy of the vehicles on a second by second basis. The energy economy is defined as the total propulsion energy used divided by the distance travelled. The energy economy is defined as the total propulsion energy used divided by the distance travelled. After the energy economy of the vehicle was computed, the next step in the analysis was to segment the data by road-type and level of congestion using a neural network. The way this was done is integral to the following Results section and detail on each step will be presented there. 3. Results and discussion The development of driving cycles that represent real-world driving conditions consists of four steps, (i) data segmentation, (ii) data classification, (iii) driving cycle synthesis and (iv) driving cycle evaluation. 3.1. Data segmentation The individual journey distance distribution is shown in Fig. 3. Fifty-five probability density functions (pdf) were fitted iteratively to the individual journey distances to find the best fit. The three parameter fatigue life distribution (Eq. (1)) was found to be the best fit for the data based on the Kolmogorov Smirnov goodness of fit test, where a = 1.2698, b = 3346.6 and c = 290.09 [50].

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffi!! ðx  cÞ=b þ b=ðx  cÞ 1 xc b f ðxÞ ¼ ;  xc a b 2aðx  cÞ

ð1Þ

In the first step, the driving cycles were segmented into four driving distance bins each having the same probability on the cumulative distribution function (cdf) (Fig. 3). A representative

169

driving cycle of the real-world cycles in each bin, four cycles in total, were developed in the study. The mean value of the bin range, indicated by a black circle in Fig. 3, was selected as a representative driving cycle distance for each bin. Only the longest cycle corresponding to bin 4 is presented here and is referred to as the Dublin driving cycle hereafter. 3.2. Data classification Driving cycles exhibited by real-world drivers are the product of the instantaneous decisions of the driver to cope with the physical driving environment. Research has shown that the driving environment of a vehicle has a strong influence on the fuel economy of the vehicle [39]. Specifically, road type, level of congestion and driving style have varying degrees of impacts on the vehicle’s fuel economy. The economy of an electric vehicle is affected not only by road type, level of congestion and driving style but also by the usage of auxiliaries [51,52] and by weather [53]. These factors can generally be observed in the velocity profile of a vehicle. In order to analyse the large amount of data collected in the form of driving cycles they were divided into micro-segments and classified as operating in one of six driving environments e.g. congested urban driving or free-flowing motorway driving. This facilitated an analysis of the energy economy of the EVs in the different driving environments and allowed the proportion of time that EVs operate in each driving environment to be identified. Abdollahi [54] developed an automated driving condition recognition algorithm based on neural networks (NN) for an adaptive power management control strategy for a parallel hybrid vehicle in order to minimise fuel consumption. Shankar et al. [55] also developed a NN methodology to determine the efficiency of a drivetrain in different driving conditions and reported that driving conditions result in distinct operating regimes in the drivetrain. The road-type and traffic condition composition of driving cycles are dependent on the journey distance and influence the energy economy of an EV over a driving cycle. Therefore, it is important that these are taken into consideration in the development of driving cycles. A driving condition recognition (DCR) tool based on a learning vector quantisation (LVQ) NN was developed to classify the driving cycle data as operating in particular driving conditions determined by road-type and level of congestion. The DCR tool allowed the proportion of time that the EVs operated in specific driving conditions in the real-world and the energy economy of the EVs in these driving conditions to be determined. The DCR tool operates by dividing each cycle into 30 s micro-segments and computing 28 parameters for each micro-segment as illustrated in Fig. 4. The

Fig. 3. Probability distribution and cumulative distribution functions of the individual driving cycle distances. The selected driving cycle distance for each bin are indicated by black circles.

170

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

Fig. 4. Illustration of the DCR tool extracting statistics from a driving cycle every 30 s.

28 parameters are listed in Table 1. Each micro-segment is then classified as a particular road-type and level of congestion by the NN. As mentioned earlier, the driving cycle dataset consisted of 1485 trips made by 7 EVs in the Dublin area over a six month period. In order to classify this dataset with the NN, it was first necessary to train the NN. The training dataset was developed by recruiting thirty different electric vehicle (EV) drivers to drive along a prescribed route (Fig. 5(left)) at three different times of the day. The route was chosen such that a variety of different road-types were encountered (i.e. urban, extra-urban and motorway). Each driver completed the route once. The three different

times of day (9:30 am, 1 pm and 4:30 pm) were chosen so that the vehicle would be driven in varying levels of congestion. The traffic congestion was heaviest during the 4:30 pm trip and least during the 9:30 am trip. The resulting thirty trips were used as a training dataset for the NN. The training dataset consisted of 898 km of driving data or 3534 micro-segments. At the end of the training process, the NN could identify six different driving environments; stop-start urban, congested urban, free-flowing urban, extra-urban, congested motorway and free-flowing motorway. The GPS data for each test route were plotted on Google Earth (Fig. 5(left)) and by examining the time stamps of the GPS data (at bottom of Fig. 5(right)) and viewing the map, it was possible to classify each micro-segment of the logged driving cycle from each test route as a particular road-type. Barth et al. [56] investigated the relationships between a number of microscopic speed fluctuation measures, which were calculated from driving cycles collected by a GPS-equipped instrumented vehicle and macroscopic traffic variables that were logged by traffic detectors on roadways. One such measure, the coefficient of variation of instantaneous speed (COVvt), was adopted as a congestion index of each micro-segment in the DCR tool. The COVvt is defined as the standard deviation of velocity divided by the mean of the velocity (expressed as percentage) of a micro segment. The boundaries of the congestion index in which the level of congestion of a micro segment could be classified as stop-start, congested or free-flowing is subjective. Montazeri-Gh and Fotouhi [57] used a k-means clustering approach in a traffic condition recognition study in the development of an intelligent hybrid electric vehicle (HEV) control strategy. The authors

Table 1 The computed micro-segment parameters. Parameter

Parameter

Parameter

Parameter

1. Average speed 2. Standard deviation of speed 9 3. Maximum speed

8. Standard deviation of deceleration 9. Maximum deceleration 10. Relative positive acceleration

22. % of time when (va) is 6–10 23. % of time when (va) is 10–15 24. % of time when (va) is >15

4. Average acceleration 5. Standard deviation of acceleration 6. Maximum acceleration

11. Number of stops per km 12. Stop duration per km

15. % of time in speed interval 30–50 km/h 16. % of time in speed interval 50–70 km/h 17. % of time in speed 18. interval 70–90 km/ h 18. % of time in speed interval 90–110 km/h 19. % of time in speed >110 km/h 20. % of time when speed ⁄ acceleration (va) <0 21. % of time when (va) is 3–6

27. Positive kinetic energy

7. Average deceleration

13. % of time in speed interval 0–15 km/ h 14. % of time in speed interval 15– 30 km/h

25. Total distance (m) 26. Total duration

28. Energy consumption per km (Wh/ km)

Fig. 5. (Left) Test route taken to generate the training dataset for the NN. (Right) An illustration of the identification of the road-type of a micro-segment.

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

171

Fig. 6. k-means clustering of the micro-segments of road-type urban.

clustered the driving cycle variables ‘idle time percentage’ and ‘average acceleration’ of 150 s micro segments into three clusters representing different traffic conditions. In the work presented here, k-means clustering was used to cluster the congestion index and average velocity of the micro-segments into three clusters representing three levels of congestion, stop-start, congested and free-flowing respectively. Donateo et al. [58] also used k-means clustering to obtain reference mini-cycles to optimise the energy management of plug-in hybrid electric vehicles. Fig. 6 illustrates the clustering of the micro segments classified as being of roadtype urban in the training dataset. A visual comparison of the classification between the trained NN output and the training data was assessed by plotting the distribution of the mean velocity against the mean acceleration for each micro segment for both the training data (Fig. 7(a)) and the NN output (Fig. 7(b)). The figures demonstrate similar patterns in terms of the distribution of velocity supported by presentation of the confusion matrix in Fig. 8 for the six target classes: stop-start urban (1), congested urban (2), free-flowing urban (3), extraurban (4), congested motorway (5) and free-flowing motorway (6). From Fig. 8, it can be seen that the NN worked well in predicting most target classes except in the case of the target class of extra urban with most confusion exhibited when trying to distinguish between free-flow urban and extra urban. In real terms, the two road conditions are quite similar both being on the outskirts of downtown urban areas. In overall terms, as can be seen from Fig. 8, the NN was successful in correctly identifying driving conditions 90% of the time. On completion of NN training, the NN was applied to the full dataset containing the logged driving data from 7 EVs over a six month period in Dublin, the classification results of which are presented in Fig. 9. The classification of the micro segment data in this way allowed the proportion of time that the vehicles spent operating in each of the driving environments to be determined. Fig. 10 presents the proportion of time spent in each driving category for the driving cycles in bin 4. The vehicles predominantly operated in urban driving conditions and approximately 7% of the time in motorway driving conditions. 3.3. Driving cycle synthesis In the third step a driving cycle model that synthesises cycles by the application of velocity and acceleration succession probabilities at a second by second level was developed. The methodology for generating the driving cycles is based on Markov process theory. When the future probabilistic behaviour of a process depends only on the present state, the resultant model is called a discrete time Markov-chain [59]. A Markov chain, with the states velocity and acceleration, has been demonstrated as being suitable to deal

Fig. 7. Comparison of (a) the training data and (b) the neural network output for the six driving environments.

with the random property of cycles [16,60,61]. The states required to satisfy the Markov property are determined based on the assumption that vehicle dynamics can be simplified using the following dynamic equation (Lee and Filipi [16].

F net ¼ F prop  F RR  F WR  F GR ¼ me av eh ¼ me v_ v eh

ð2Þ

where Fnet is the net force applied to the vehicle, Fprop is the propulsion force from the powertrain, FRR is the rolling resistance force, FWR is the wind resistance force, FGR is the grade resistance force and all other external forces applied to the vehicle, me is the equivalent vehicle mass, vveh is the vehicle velocity and aveh is the vehicle acceleration. In Eq. (2), vehicle dynamics can be represented using two states; vehicle velocity and acceleration. They are the states

172

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

Fig. 8. Confusion matrix for neural network training data.

Fig. 9. The distribution of velocity and acceleration for the entire dataset.

3.4. Driving cycle evaluation

Fig. 10. Driving environment composition of the driving cycles in bin 4.

selected for the Markov chain. Velocity and acceleration data of the selected driving cycle are gathered and then a transition probability matrix (TPM) is generated at the current time tk. In each cell of the TPM, the probability matrix of the velocity and acceleration at the next time tk+1 are included. Each cell represents the transitional probability Pðv kþ1 ¼ v 2 ; akþ1 ¼ a2 jv k ¼ 1; ak ¼ 1Þ at a given cycle velocity and acceleration [16].

Statistical analysis was necessary to determine the least number of statistically significant parameters that influence the energy economy of a vehicle over a driving cycle. A similar regression analysis was performed to that proposed by Lee and Filipi [16] and used by Önnegren [62] for assessing the representativeness of developed driving cycles for PHEVs in Michigan and ICEVs in Gothenburg respectively. The explanatory variables in the final regression equation act as the statistical criteria for the assessment of the representativeness of a synthesised candidate driving cycle. The value of the variables of a candidate cycle was required to match the mean value of the variables of the real-world driving cycles within ±10% in order to guarantee that a particular synthesised cycle satisfactorily represents the logged real-world driving cycles in a particular bin. The response variable in the regression analysis was the energy economy of the vehicle over a driving cycle. It was defined as the total energy used by the vehicle divided by the distance travelled (W h/km). This parameter has been used as a representative response variable in previous driving cycle studies [16,27,63]. Initially, 27 explanatory variables were selected and categorised into velocity related, acceleration related, driving time and distance related, and driving characteristics related groups as shown in Table 2. However, only a selection of those variables were nominated as initial explanatory variables for the regression analysis through investigations of the correlations between each variable. Firstly, the Pearson product-moment correlations between the

173

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178 Table 2 The twenty-seven explanatory variables, the fifteen initial nominated variables and the eleven final nominated explanatory variables for the regression analysis. Category

Initial possible explanatory variables

Initial nominated explanatory variables

Final nominated explanatory variables

1. Velocity related variables

1. 2. 3. 4.

1. Mean velocity

1. Mean velocity

2. Standard deviation of velocity

2. Standard deviation of velocity

3. Mean positive acceleration

3. Mean positive acceleration

2. Acceleration related variables

Mean velocity Maximum velocity Standard deviation of velocity Mean cruising velocity

5. Mean positive acceleration 6. Mean negative acceleration 7. Positive acceleration time 8. Negative acceleration time 9. Maximum acceleration 10. Minimum acceleration 11. Standard deviation of acceleration 12. Standard deviation of positive acceleration 13. Standard deviation of negative acceleration 14. Percentage of driving time under positive acceleration 15. Percentage of driving time under negative acceleration

3. Driving distance and time

16. Driving distance 17. Driving time 18. Total time

4. Driving characteristics

19. 20. 21. 22.

Idle time Percentage of idle time Number of stops Number of stops per kilometre

23. 24. 25. 26. 27.

Mean specific power Maximum specific power Minimum specific power Cruising time Percentage of cruising time

variables within the four categories were investigated. When two variables demonstrated a high correlation, the variable which was least correlated with the response variable was excluded as an explanatory variable. In Fig. 11(a), an example of a within group correlation is presented; the standard deviation of velocity is highly correlated with the maximum velocity. Thus, one of these variables could be excluded from the analysis. This process resulted in fifteen initial nominated explanatory variables as shown in Table 2. Secondly, the correlations between the fifteen variables (i.e. between the groups) were investigated. Similarly, when two variables demonstrated a high correlation, one of them was excluded from the initial set of fifteen nominated variables. In Fig. 11(b), an example of a between group correlation is presented; mean power is highly correlated with average velocity. Thus, one

4. Positive acceleration time 5. Maximum acceleration 6. Minimum acceleration

4. Maximum acceleration 5. Minimum acceleration

7. Percentage of driving time under negative acceleration 8. Driving distance

9. Percentage of idle time 10. Number of stops 11. Number of stops per kilometre

6. Percentage of idle time 7. Number of stops 8. Number of stops per kilometre

12. Mean specific power 13. Maximum specific power

9. Maximum specific power

14. Cruising time 15. Percentage of cruising time

10.Cruising time 11. Percentage of cruising time

Table 3 The determined significant explanatory variables and the final regression equation. Predictor

Coefficient

Standard error of coefficient

P

Constant Percentage cruise time Maximum specific power Mean positive acceleration Numbers of stops per km Mean velocity Cruise time Minimum acceleration Number of stops Maximum acceleration Percentage of Idle time

99.102 1.048 0.00049 76.308

2.922 0.054 0.00003 4.045

0.000 0.000 0.000 0.000

4.646 0.606 0.023 4.288 0.263 3.612 0.078

0.185 0.048 0.002 0.564 0.048 0.800 0.035

0.000 0.000 0.000 0.000 0.000 0.000 0.025

R2 = 82.7%: Adjusted R2 = 82.5%.

Fig. 11. (a) The correlation between two within-group explanatory variables: between the standard deviation of velocity and maximum velocity. (b) The correlation between two between group nominated explanatory variables: between average velocity and mean power.

174

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

Fig. 12. (a) Normal probability plot of standardised residuals (b) histogram of the standardised residuals.

of these variables could be excluded. This process reduced the fifteen initial nominated variables to the final eleven nominated variables to be included in the regression analysis as shown in Table 2. Stepwise regression, a semi-automated process of building a model by successively adding or removing variables based solely on the t-statistics of the estimated coefficients, was used to choose the explanatory variables from the eleven initial explanatory variables to be included in the model. This method ensured that the model contained the smallest possible set of predictor variables. Ten of the final eleven nominated explanatory variables were included in the final model. To obtain the final regression equation a multiple regression analysis was performed. The resulting model was statistically significant and accounted for approximately 82.5% of the variance of the response variable, energy economy (R2 = 82.7%, Adjusted R2 = 82.5%). The final regression equation consisted of ten variables as shown in Table 3. The validity of the developed regression equation can be assessed by examining the normal probability plots (Fig. 12(a)) and histogram of the residual (Fig. 12(b)). Fig. 13 illustrates the predicted energy economy using the developed regression equation versus the actual energy economy of the recorded real-world driving cycles. The plot indicates that the predicted values are well correlated with the real-world values. Although the points are not perfectly aligned on the diagonal, the distribution is relatively tight, indicating that the regression equation captures the main characteristics of driving cycles that influence the energy economy of an EV.

3.5. The Dublin city driving cycle The Markov-chain process was performed in order to satisfy the ten statistical variables determined by the regression analysis. The values of the variables of the candidate cycles were required to match the mean value of the variables of real-world driving cycles within ±10%; a range the authors considered to be a reasonable degree of deviation. The Dublin city EV driving cycle is presented in Fig. 14. The statistical parameters of the developed cycle are shown to represent well the mean values of the variables of the real-world driving cycles as shown in Table 4.

Fig. 14. Dublin city driving cycle.

Table 4 Comparison of the statistical parameters of the Dublin city EV cycle with the parameters of the real-world driving cycles.

Fig. 13. Comparison of the real-world energy economy values versus the predicted energy economy by the regression equation.

a

Statistical parameters

Real-world driving cyclesa

The Dublin city EV cycle

Average acceleration (m/s2) Number of stops per km Percentage of cruise time (%) Maximum specific power (W/km) Number of stops Minimum acceleration (m/s2) Percentage of idle time (%) Average velocity (km/h) Max acceleration (m/s2) Cruising time (s)

0.59 1.4 25.95 4540 20 2.56 19.64 30.60 2.3 495

0.62 1.42 27.6 4.805 20 2.5 20.67 30.87 2.22 454

Mean values are presented for the real-world driving cycles.

175

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

Fig. 15. Probability distribution (blue bars) and cumulative distribution (red line) of the energy economy of the real-world driving cycles and the predicted energy economy of the Dublin city cycle (green). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

comparison because one could not expect ICEV and EV driving cycles to be similar given the significant differences in vehicle technologies but it perhaps can put in context the results of the work against other driving cycle research. Zaccardi and Le Berr [65] list 18 worldwide legislative and non-legislative cycles from Barlow et al. [66] suitable for the homologation and design of EVs. From that list, those with a focus on urban conditions were selected along with the Winnipeg cycle, developed by Ashtari et al. [36] for EVs in the city of Winnipeg, were used to put in context the Dublin EV cycle (see Table 5). The method used analysed the difference between the Speed Acceleration Frequency Distribution (SAFD) of each of the cycles (both the developed and the well-established cycles) and all of the recorded driving cycle data. A SAFD expresses the amount of time spent in specific speed and acceleration bins. This technique has been used previously to evaluate developed cycles [36,67,68]. The SAFD of the Dublin city EV driving cycle is illustrated in Fig. 17. Fig. 16. Dublin city driving cycle driving condition proportions.

The energy economy (blue1 bars) of the real-world driving cycles are distributed broadly as illustrated in Fig. 14. The distribution is overlaid with a cdf (red line). Tae-Kyung and Filipi [64] assumed that if the predicted energy economy of the developed cycle is located around the median point of the observed real-world energy economy values then the developed driving cycle (shown in green) represents well the recorded driving cycles. Fig. 15 illustrates that the developed cycle is representative in terms of the prediction of the energy economy of an EV over the cycle. Finally, the developed cycle was assessed to ensure that it consists of the same proportions of driving conditions in terms of road types and traffic conditions to those observed in real-world operating conditions. Fig. 16 illustrates that the developed cycle consists of approximately the same proportions of driving conditions to those observed in the real-world driving cycles.

Table 5 A selection of other driving cycles. Region

Driving cycles

Description

Type

European cycles

NEDC Artemis traffic jam Artemis urban Artemis road

Composed of ECE & EUDC Congested urban traffic

Modal Transient

Urban traffic Road conditions

Transient Transient

Hyzem urban Eurev UL1 Eurev UF1

Urban traffic Congested urban traffic Urban traffic (moderate speed) Urban traffic (high speed) Extra-urban traffic (moderate speed) Extra-urban traffic (high speed) Motorway traffic (moderate speed)

Transient Transient Transient

French cycles

Eurev UF3 Eurev R1 Eurev R3 Eurev A1

3.6. The EV driving cycle in the context of existing ICEV driving cycles In the final step, the developed cycle is presented alongside a selection of well-established driving cycles. This is not strictly a 1 For interpretation of color in Fig. 14, the reader is referred to the web version of this article.

Transient Transient Transient Transient

Japanese cycle

Japanese 10–15 Mode

Urban traffic (high speed)

Modal

US cycles

FTP 75 LA92 New York city

Urban traffic (high speed) Extra-urban traffic Urban traffic

Transient Transient Transient

Winnipeg, Canada

Winnipeg cycle

Real world EV cycle

Transient

176

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

Fig. 17. The speed acceleration frequency distribution of the Dublin city driving cycle.

The smaller the sum of differences between the SAFDs, the higher the commonality between the cycle in question and the logged data in a bin. SAFDdiff represents the percentage difference between the SAFD of all the data and a selected cycle as defined by Eq. (3).

P SAFDdiff ¼

i ðSAFDcycle ðiÞ

P

2

 SAFDdata ðiÞÞ

i SAFDdata ðiÞ

ð3Þ

2

where i is the ith bin in the SAFD, SAFDcycle is the SAFD of a cycle, and SAFDdata is the SAFD of all the data. The abbreviations of the variables to be compared are listed in Table 6. Table 7 presents the values of the variables of the devel-

Table 6 Final driving cycle variables, abbreviations and units. Variable

Abbreviation

Unit

Percentage cruising time Mean positive acceleration Number of stops per kilometre Mean velocity Minimum acceleration Maximum acceleration Percentage of idle time

P_cru Acc_pos Stop_per_km V_avg Acc_min Acc_max P_idle

s m/s2 NA km/h m/s2 m/s2 %

oped cycle and the 16 other cycles and presents the SAFDdiff between them to the recorded real-world cycles. The results in Table 7 shows that the characteristics of the Dublin city EV cycle are close to the mean value of the variables of the real-world cycles. The SAFDdiff between the real-world cycles and the Dublin city cycle is 4%, which is the joint lowest value along with the Winnipeg EV cycle. The Artemis Urban, the Hyzem Urban and the FTP-75 cycles are the next closest cycles in terms of minimising the SAFDdiff, with values in the range of 6–8%. The least similar cycles in terms of matching the mean of the statistical parameters of the real-world cycles are the Eurev UF1, the Artemis Traffic Jam, and the Eurev UL1 cycles. To illustrate, the Eurev UF1 cycle has a mean velocity of 10.5 km/hr compared to 30.6 km/hr, the mean velocity of the real-world cycles. The cycle also has a large number of stops per kilometre, 6.88 compared to 1.4, the mean number of stops per kilometre of the real-world cycles. Similarly, the mean velocity and number of stops per kilometre of the Eurev UL1 and the Artemis Traffic Jam cycles, 3.9 m/h and 42.12 respectively and 10.6 km/h and 9.87 respectively, are dissimilar to the mean velocity and mean number of stops per kilometre of the recorded real-world cycles. However, only the Artemis Traffic Jam cycle has a large SAFDdiff value of 57%. The SAFDdiff values of the Eurvev UF1 and Eurev UL1 cycles are 16% and 28% respectively. The differences between the developed cycles and the existing legislative cycles are undoubtedly due to the fact that the data were collected from EVs for the developed cycles and from ICEVs for most of the other cycles, except in the case of the Winnipeg EV cycle. An electric motor can provide maximum torque to the wheels immediately and can preserve this until the vehicle reaches the speed of maximum power. In contrast, an ICEV must work up to the maximum torque through gear shifts. In addition, the EVs tend to operate mainly in urban areas where the speed limits are typically 50 km/h and lower extending up to 80 km/h on the outskirts of urban areas. The two characteristics of large average accelerations and low average velocities which are present in the data can be attributed to these facts. The results from this work can be used for a number of real world applications. Some of the recent research into driving cycles for ICEVs is focused on vehicle emission estimation [69–71]. When comparing the emissions of different vehicle technologies it is important to have drive cycles that are representative of the real world driving conditions and environment. The driving cycle presented here for EVs could be used in environmental impact evaluation comparisons particularly when estimating the emissions of the associated electricity production. Wang et al. [15] point to

Table 7 Statistical parameters of the Dublin city driving cycle and sixteen other cycles.

⁄

Real-world cycles Dublin city cycle FTP-75 Hyzem urban Eurev UF3 NEDC Winnepeg cycle Japanese 10–15 mode LA92 Eurev R1 Eurev R3 Artemis road Artemis urban Eurev A1 New York city cycle Eurev UF1 Artemis traffic jam Eurev UL1

Acc_pos (s)

Stops_per_km

V_avg (km/h)

Acc_min (m/s2)

Acc_max (m/s2)

P_idle (%)

P_cru (%)

SAFDdif (%)

0.59 0.62 0.65 0.71 0.73 0.60 0.68 0.52 0.74 0.99 0.52 0.58 0.78 0.52 0.75 0.71 0.88 0.64

1.4 1.42 1.33 1.15 1.52 1.09 1.3 1.10 0.95 1.15 0.06 0.17 4.31 0.07 6.33 6.88 9.87 3.9

30.6 30.87 32.1 23.0 25.1 34.5 32.9 28.0 40.7 32.0 58.7 58.0 18.1 75.2 13.2 10.5 10.6 42.12

2.56 2.5 1.67 2.06 3.89 1.39 6.39 0.83 4.17 4.17 2.50 4.17 3.05 3.06 2.78 2.22 2.5 2.5

2.3 2.22 1.67 2.19 2.22 1.11 2.78 0.81 3.06 2.78 2.22 2.5 2.78 3.06 2.78 2.5 2.22 1.94

19.64 20.67 17.7 22.4 12.43 22.8 20.3 26.4 14.0 16.6 7.3 2.24 26.7 1.4 30.0 30.1 40.7 36.57

26.0 27.6 25.3 8.8 14.64 41.1 22.1 22.8 20.47 9.5 29.0 29.45 10.0 36.1 11.4 11.0 7.3 15.0

4 8 6 26 66 4 33 19 12 51 85 6 110 22 16 57 28

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

the importance of developing and using real-world driving cycles in designing and evaluating electric vehicles. Using the driving cycle presented in this paper would be much more useful and effective when designing and evaluating electric vehicles than using driving cycles derived from ICEVs given the very large differences in drive trains and the differences in accelerations and velocities between the two vehicle types, as mentioned earlier. Allied to this would be the use of the driving cycle for better estimation of battery range; a key aspect of design, marketing of EVs and providing more accurate information to EV users.

177

Appendix A (continued)

Parameter Maximum current of the motor Motor mass Efficiency Mass of the torque coupling Torque coupling ratio Mass per wheel Radius of wheel Maximum braking torque of the vehicle

616.4 A 64.2 kg 95% 10 kg 1.6 30 kg 0.317 2000 Nm

4. Conclusions A driving cycle was developed using a database of driving data from a fleet of EVs over a six month period with velocity recorded on a second by second basis in Dublin, Ireland. The developed driving cycle was shown to consist of the same proportions of driving conditions in terms of road types and traffic conditions to those observed in real-world operating conditions. In addition, the predicted energy economy of the developed driving cycle was shown to be representative of real-world energy economy values. Real-world driving patterns differ significantly from legislative cycles used for design and certification in the United State of America, Europe, and Japan. They generally consist of lower velocity driving and higher acceleration rates. This implies that these cycles may not be suitable for design purposes of EVs for real-world applications. Real-world driving cycles are essential for EV powertrain design, battery management systems, battery range estimation and the provision of better information to EV users. The developed driving cycle would aid in the design of EVs that are operating, not just in Dublin, but in urban areas in other medium sized cities. In addition, the developed driving cycle would allow electricity grid analysis, economic and lifecycle studies to be conducted with a higher degree of confidence. Acknowledgements The authors would like to thank the Higher Education Authority Programme for Research in Third Level Institutions (PRTLI IV) for funding this research. The authors would also like to the Electricity Supply Board (ESB), for providing the driving cycle dataset. Appendix A. Powertrain model specification for the BEV

Parameter 12 V battery mass Constant power loss Vehicle body mass Vehicle cargo mass Vehicle centre of gravity Coefficient of drag Frontal area of vehicle Number of cells in the ultra-capacitor pack connected in parallel Number of cells connected in series Max capacity of the energy storage system Max Power of the energy storage system Final drive ratio Coefficient of regen for the motor Motor controller mass

18 kg 200 W 1525 kg 136 kg 0.5 m 0.29 2.27 m2 2 96 66 A h 468 W 7.94 1 25.6 kg

References [1] Hill G, Blythe PT, Higgins C. Deviations in Markov chain modeled electric vehicle charging patterns from real world data. In: Procs of 15th international IEEE conference on intelligent transportation systems (ITSC), 2012, Anchorage; 16–19 September 2012. p. 1072–77 [2] Donateo T, Licci F, D’Elia A, Colangelo G, Laforgia D, Ciancarelli F. Evaluation of emissions of CO2 and air pollutants from electric vehicles. Appl Energy 2015;157:675–87. [3] Onat NC, Kucukver M, Tatari O. Conventional, hybrid, plug-in hybrid or electric vehicles? State based comparative carbon and energy footprint analysis in the US. Appl Energy 2015;150:36–49. [4] Mallouh MA, Abdelhafez E, Salah M, Hamden M, Surgenor B, Youssef M. Model development and analysis of a mid-sized hybrid fuel cell/battery vehicle with a representative driving cycle. J Power Sources 2014;260:62–71. [5] Saxena S, MacDonald J, Moura S. Charging ahead on the transition to EVs with standard 120 V wall outlets. Appl Energy 2015;157:720–8. [6] Weldon P, Morrissey P, Brady J, O’Mahony M. An investigation into usage patterns of electric vehicles in Ireland. Transport Res Part D: Transp Environ 2016;43:207–25. [7] Wikstron M, Hansson L, Alvfors P. An end has a start – investigating the usage of electric vehicles in commercial fleets. In: Procs of the 7th international conference on applied energy, ICAE2015. Energy Procedia, vol. 75; 2015, p. 1932–37. [8] Morrissey P, Weldon P, O’Mahony M. Future standard and fast charging infrastructure planning: an analysis of electric vehicle charging behaviour. Energy Policy 2016;89:257–70. [9] Weldon P, Morrissey P, O’Mahony M. Environmental impacts of varying electric vehicle user behaviours and comparisons to internal combustion engine vehicle usage – an Irish case study. J Power Sources 2016;319:27–38. [10] Meng J, Mu Y, Jia H, Wu J, Yu X, Qu B. Dynamic frequency response from electric vehicles considering travelling behavior in the Great Britain power system. Appl Energy 2016;162:966–79. [11] Schill W, Gerbaulet C. Power system impacts of electric vehicles in Germany; charging with coal and renewables. Appl Energy 2015;156:185–96. [12] Saxena S, Gopal A, Phadke SA. Electrical consumption of two-, three- and fourwheel light-duty electric vehicles in India. Appl Energy 2014;115:582–90. [13] Seedam A, Satiennam T, Radpukdee T, Satiennam W. IATSS Res 2015;39:79–85. [14] Rangaraju S, de Vroey L, Messagie M, Mertens J, van Mierlo J. Impacts of electricity mix, charging profile and driving behaviour on the emissions performance of battery electric vehicles – a Belgian case study. Appl Energy 2015;148:496–505. [15] Wang H, Zhang X, Ouyang M. Energy consumption of electric vehicles based on real-world driving patterns: a case study of Beijing. Appl Energy 2015;157:710–9. [16] Lee T-K, Filipi ZS. Synthesis of real-world driving cycles using stochastic process and statistical methodology. Int J Veh Des 2011;57:17–36. [17] Kruse R, Huls T. Development for the federal urban driving cycle. SAE Technical Paper. 730553; 1973. [18] Dieselnet. Emission test cycles ECE 15 + EUDC/NEDC. Last updated 07/2014. Available: ; 2013 [accessed 27.01.16]. [19] Hung WT, Tong HY, Lee CP, Ha K, Pao LY. Development of a practical driving cycle construction methodology: a case study in Hong Kong. Transport Res Part D: Transp Environ 2007;12:115–28. [20] Esteves-Booth A, Muneer T, Kirby H, Kubie J, Hunter J. The measurement of vehicular driving cycle within the city of Edinburgh. Transport Res Part D: Transp Environ 2001;6(3):209–20. [21] Tzirakis E, Pitsas K, Zannikos F, Stournas S. Vehicle emissions and driving cycles: comparison of the Athens driving cycle (ADC) with ECE-15 and European driving cycle (EDC). Global NEST 2006;8(3):282–90. [22] Amirjamshidi G, Roorda MJ. Development of simulated driving cycles for light, medium and heavy duty trucks: case of the Toronto Waterfront area. Transp Res Part D 2015;34:255–66.

178

J. Brady, M. O’Mahony / Applied Energy 177 (2016) 165–178

[23] Ho S, Wong Y, Change VW. Developing Singapore Driving Cycle for passenger cars to estimate fuel consumption and vehicular emissions. Atmos Environ 2014;97:353–62. [24] Ericsson E. Variability in urban driving patterns. Transport Res Part D: Transp Environ 2000;5(5):337–54. [25] Seers P, Nachin G, Glaus M. Development of two driving cycles for utility vehicles. Transp Res Part D 2015;41:377–85. [26] Helmers E, Marx P. Electric cars: technical characteristics and environmental impacts. Environ Sci Eur 2012;24:1–15. [27] Fellah M, Singh G, Rousseau A, Pagerit S, Nam E, Hoffman G. Impact of realworld drive cycles on PHEV battery requirements. SAE technical paper 200901-1383; 2009. [28] Tate ED, Harpster M, Savagian P. The electrification of the automobile: from conventional hybrid, to plug-in hybrids, to extended-range electric vehicles. In: 2008 World Congress. Detroit, Michigan: SAE Technical Paper 2008-010458; 2008. [29] EPA. Emission test cycles EPA Highway Fuel Economy Test Cycle (HWFET). Last updated 2/6/2016. Available: ; 2013 [accessed 27.01.16]. [30] Patil R, Adornato B, Filipi ZS. Impact of naturalistic driving patterns on PHEV performance and system design. SAE Technical Paper 2009-01-2715; 2009. [31] Kang S, Min K. Dynamic simulation of a fuel cell hybrid vehicle during the federal test procedure-75 driving cycle. Appl Energy 2016;161:181–96. [32] Gonder J, Markel T, Thornton M, Simpson A. Using global positioning system travel data to assess real-world energy use of plug-in hybrid electric vehicles. Transport Res Rec: J Transport Res Board 2007;2017:26–32. [33] Shahidinejad S, Bibeau E, Filizadeh S. Statistical development of a duty cycle for plug-in vehicles in a North American Urban setting using fleet information. IEEE Trans Veh Technol 2010;59(8):3710–9. [34] Karabasoglu O, Michalek J. Influence of driving patterns on life cycle cost and emissions of hybrid and plug-in electric vehicle powertrains. Energy Policy 2013;60:445–61. http://dx.doi.org/10.1016/j.enpol.2013.03.047. September 2013. [35] Rask E, Rousseau A. Evaluating real world drive cycles to support APRF technology evaluations. In: 2012 DOE hydrogen program and vehicle technologies. Argonne National Laboratory. US Dept of Energy. ; 2012 [accessed 27.01.16]. [36] Ashtari A, Bibeau E, Shahidinejad S. Using large driving record samples and a stochastic approach for real-world driving cycle construction: winnipeg driving cycle. Transport Sci 2012;48(2):170–83. [37] Kwon J, Kim J, Fallas E, Pagerit S, Rousseau A. Impact of drive cycles and powertrain configurations on PHEV battery component requirements. SAE World Congress and Exhibition. Detroit. Technical Paper. 2008-01-1337; 2008. [38] Dubarry M, Svoboda V, Hwu R, Liaw BY. A roadmap to understand battery performance in electric and hybrid vehicle operation. J Power Sources 2007;174(2):366–72. [39] Ericsson E. Independent driving pattern factors and their influence on fuel-use and exhaust emission factors. Transport Res Part D: Transp Environ 2001;6 (5):325–45. [40] Watson H. Effects of a wide range of drive cycles on the emissions from vehicles of three levels of technology. SAE Technical Paper 950221; 1995. [41] Rakha H, Dion F, Sin H-G. Field evaluation of energy and emission impacts of traffic flow improvement projects using GPS data: issues and proposed solutions. Transport Res Rec: J Transport Res Board 2001;1768:210–23. [42] Hellinga B. Issues related to quantifying the environmental impacts of transportation strategies using GPS data Available:Available from: 2011 [accessed 27.01.16]. [43] Riches E. Track tested: 2010 Mitsubishi i MiEV vs. Mitsubishi i [Online]. edmunds.com. Available: ; 2010 [accessed 27.01.16]. [44] Autonomie. Autonomie. Available: ; 2013 [accessed 27.01.16]. [45] Kim N, Rousseau A, Rask E. Autonomie model validation with test data for 2010 Toyota Prius. SAE 2012-01-1040, SAE World Congress, Detroit, MI; 2012 April.

[46] Kim N, Carlson R, Jehlik F, Rousseau A. Tahoe HEV Model Development in Powertrain System Analysis Toolkit (PSAT). SAE paper 2009-01-1307, SAE World Congress, Detroit, MI; 2009 April. [47] Cao Q, Pagerit S, Carlson R, Rousseau A. PHEV hymotion prius model validation and control improvements. In: 23rd international electric vehicle symposium (EVS23), Anaheim, CA, (December 2007). [48] Rousseau A, Sharer P, Pagerit S, Duoba M. Integrating data, performing quality assurance, and validating the vehicle model for the 2004 Prius Using PSAT. SAE paper 2006-01-0667, SAE World Congress, Detroit, MI; 2006 April. [49] Nissan, Ireland. ; 2016 [accessed 11.05.16]. [50] Leiva V. The Birnbaum-Saunders distribution. Elsevier Academic Press; 2015. ISBN: 9780128037690. [51] Donateo T, Ingrosso F, Licci F, Laforgia D. A method to estimate the environmental impact of an electric city car during six months of testing in an Italian city. J Power Sources 2014;270:487–98. [52] Donateo T, Ingrosso F, Bruno D, Laforgia D. Effect of driving conditions and auxiliaries on mileage and CO2 emissions of a gasoline and an electric city car. SAE Technical Paper; 2014. [53] Jayakumar A, Ingrosso F, Rizzoni G, Meyer J, Doering J. Crowd sourced energy estimation in connected vehicles IEEE International Electric Vehicle Conference, IEVC 2014, art. no. 7056189; 2014. [54] Abdollahi AD. An intelligent control strategy in a parallel hybrid vehicle. In: IEEE conference on electric and hybrid vehicles, 2006. ICEHV ’06, Pune; 18–20 December 2006. p. 1–2. [55] Shankar R, Marco J, Assadian F. A methodology to determine drivetrain efficiency based on external environment. In: 2012 IEEE international conference on electric vehicle (IEVC), Greenville. 4–8 March. p. 1–6. [56] Barth M, Johnston E, Tadi R. Using GPS technology to relate macroscopic and microscopic traffic parameters. Transport Res Rec: J Transport Res Board 1996;1520:89–96. [57] Montazeri-Gh M, Fotouhi A. Traffic condition recognition using the -means clustering method. Sci Iranica 2011;18(4):930–7. [58] Donateo T, Pacella D, Laforgia D. A method for the prediction of future driving conditions and for the energy management optimisation of a hybrid electric vehicle (2012). Int J Veh Des 2012;58(2–4):111–33. [59] Jensen P, Bard J. Operations research models and methods. John Wiley & Sons; 2002. [60] Alessandrini A, Orecchini F. A driving cycle for electrically-driven vehicles in Rome. Proc Inst Mech Eng, Part D: J Autom Eng 2003;217(9):781–9. [61] Gong Q, Midlam-Mohler S, Marano V, Rizzoni G. An iterative Markov chain approach for generating vehicle driving cycles. SAE Int J Eng 2011;4:1035–45. [62] Önnegren P. Driving cycle generation using statistical analysis and Markov chains. Linköpings universitet. Masters Thesis. ; 2013 [accessed 27.01.16]. [63] Carlson TR, Austin TC. Development of speed correction cycles. Sacramento, California: Sierra Research, Inc.; 1997. Report No. SR97-04-01. [64] Tae-Kyung L, Filipi ZS. Synthesis and validation of representative real-world driving cycles for Plug-In Hybrid vehicles. In: 2010 IEEE vehicle power and propulsion conference (VPPC), IEEE, Lille, 1–3 September 1–6. [65] Zaccardi J, Le Berr F. Analysis and choice of representative drive cycles for light duty vehicles – case study for electric vehicles. Proc Inst Mech Eng, Part D: J Autom Eng 2012;227(4):605–16. [66] Barlow T, Latham S, McCrae I, Boulter P. A reference book of driving cycles for use in the measurement of road vehcile emissions. Project Report PPR354, TRL. ; 2009. [67] Bata R, Yacoub Y, Wang Wg, Lyons D, Gambino M, Rideout G. Heavy duty testing cycles: survey and comparison. International truck and bus meeting and exposition. Seattle. SAE Technical Paper 942263; 1994. [68] Lin J, Niemeier D. An exploratory analysis comparing a stochastic driving cycle to California’s regulatory cycle. Atmos Environ 2002;36:5759–70. [69] Knez M, Muneer T, Jereb B, Cullinane K. The estimation of a driving cycle for Celje and a comparison to other European cities. Sustain Cities Soc 2014;11:56–60. [70] Achour H, Olabi AG. Driving cycle developments and their impacts on energy consumption of transportation. J Clean Prod 2016;112:1778–88. [71] Marques S, Batista T, Silva C. Eco-rating methodologies for private cars: driving cycle influence. Procedia – Soc Behav Sci 2014;111:682–91.