ceur-ws.org/Vol-2399/paper06.pdf Pre-Estimation of Electric Vehicle Energy Consumption on Unfamiliar Roads and Actual Driving Experiments Toshiaki Uemura supervised by Takashi Tomii Yokohama National University uemura-toshiaki-xh@ynu.jp ABSTRACT As described in this report, we propose a system that sup- For this study, we constructed a system for pre-estimation ports an EV user’s travel planning on unfamiliar roads. We of electric vehicle (EV) energy consumption on unfamiliar present a solution for pre-estimating the EV energy con- roads. Drivers of EVs must make plans based on estimated sumption range: minimum energy consumption Emin and energy consumption because they fear that an EV might run maximum energy consumption Emax . We present Emin and out of power and stop on the road. Our constructed system Emax to assist planning. In addition, if Emin and Emax are pre-estimates ranges at which an EV can be expected to correct, actual energy consumption Ereal is in the range of be forced to stop on a road. The range is calculated using Emin – Emax . Therefore, to evaluate the accuracy of the EV driving simulation on a route that is input by a driver. proposed system, we conducted EV driving experiments on The driving simulation requires only map data and the EV roads with two conditions and confirmed that Ereal is in the car specifications. Moreover, we assessed a system using a range of Emin – Emax . real EV. Results show that the system produced sufficiently correct ranges on the roads used for experimentation. Ad- 2. RELATED WORKS ditionally, we evaluated the accuracy of ranges output by Studies of many types have estimated EV energy con- our system. For evaluation, we used numerous accumulated sumption and therefore the EV travel range. Using a mo- daily driving logs for EVs. tion equation model and actual driving logs collected by a probe car, most of these studies have produced methods to calculate EV energy consumption or travel range. Grubwin- 1. INTRODUCTION kler et al. estimated EV energy consumption from statistical In recent years, energy-efficiency and CO2 emission re- analysis of driving data generated from large amounts of col- duction have become necessary worldwide because of cli- lected driving data[5]. Ito et al. estimated EV travel ranges matic variation and scarcity of fossil fuels. Given that back- from averaging energy consumption maps from a probe car ground, electric vehicles (EVs) are attracting global atten- database[6]. Zhang et al. proposed estimation of EV travel tion. Reportedly, EVs present the benefit that so-called range using driving logs, traffic conditions, and weather[12]. well-to-wheels CO2 emissions are lower than those of in- Styler et al. proposed a means of controlling a Range EX- ternal combustion vehicles (ICVs). In addition, EVs have tender (REX) EV more efficiently using estimated energy no emissions when they are running. Many countries have consumption generated from probe car data[9]. Yang et al. formulated EV deployment goals for the future. Therefore, proposed a means of estimating energy consumption and EVs are expected to penetrate markets gradually worldwide. CO2 emissions from average speed and stop frequency data Nevertheless, many difficulties arise when a user operates acquired by passage sensors at an intersection[11]. an EV. One is the difficulty of EV travel planning when Moreover, many studies have solved optimization prob- a user navigates unfamiliar roads. Planning must be done lems of energy consumption and driving using motion equa- while considering an EV travel range and when and where tions and other data. Karbowski et al. proposed a means one might stop at a charging station. However, EV travel of controlling plug-in hybrid EVs (PHEVs) using an energy ranges change drastically because of road gradients and traf- consumption simulation generated from traffic, road maps, fic conditions. Therefore, average users have difficulty mak- and Markov Chain[7]. Kurtulus and Inalhan proposed a ing a precise plan for unfamiliar routes. route decision algorithm for REXEV considering energy con- sumption calculated from traffic, weather, maps, and the destination[8]. De Souza et al. proposed a traffic assign- ment algorithm that minimizes EV travel time and energy consumption[1]. Felipe et al. estimated energy consump- tion using an artificial neural network into which driving styles and route features are input[4]. Fei et al. proposed hybrid models incorporating a motion equation model and a machine learning model[3]. Unlike these studies, we make Proceedings of the VLDB 2019 PhD Workshop, August 26th, 2019. Los our contribution by evaluating the practicality of our system Angeles, California. Copyright (C) 2019 for this paper by its authors. using large amounts of data acquired in di↵erent regions. Copying permitted for private and academic purposes. Table 1: Variables of Expression (1) Table 2: Experiment Trips Variable Unit Meaning How to get Route ID Distance Type Charging Spot gravitational kitami 1 163 km Long 121 km point g m/s2 constant acceleration kitami 2 126 km Long 67 km point ⇢ kg/m3 air density constant 85 km point rolling kitami 10 163 km Long 121 km point µ - resistance constant outward 27 km Short do not care coefficient homeward 24 km Short do not care air resistance car Cd - coefficient specification face area of car A m2 the vehicle specification mass of car M kg the vehicle specification inertia mass of car Mi kg the vehicle specification ↵ rad road gradient map data constant v m/s velocity speed is set conversion car ⌘ - efficiency specification 3. PRE-ESTIMATION SYSTEM To pre-estimate EV energy consumption, the EV user in- puts only an origin and a destination and anticipated stop locations (sightseeing spots or stores, etc.) during a trip. Next, the system generates origin-destination (OD) trip sim- ulation logs running on candidate routes at a constant speed Figure 1: Altitude and Distance of Experiment vc from these inputs. Trip simulation logs are normalized Trips. by time. We set speed vc in advance, for example, a speed limit on a road. accelerates and decelerates when stopping at every stop lo- Then, trip simulation logs are input to an EV energy con- cation and every signal, assuming bad conversion efficiency sumption model. We use a model based on a motion equa- during acceleration and deceleration.” tion[2]. Then Emin and Emax are calculated from outputs Therefore, we define Emax as shown in Expression (3) of the EV energy consumption model. because we want to express it easily. Eacc is described in Expression (4). Additionally, N stands for the number of 3.1 Emin Calculation stops when an EV stops at every stop location and every signal. This subsection presents a description of minimum energy Emin calculation. For this report, Emin is defined as “energy consumption when an EV runs at constant speed vc and does Emax = Emin + Eacc [kW h] (3) not stop.” First, an EV energy consumption log is calculated every second by inputting a trip simulation log into the EV en- 1 1 Eacc = N ⇥ (M + Mi )vc2 ⇥ [kW h] (4) ergy consumption model. Expression (1) represents the EV 2 ⌘ energy consumption model. Table 1 presents variables of In Eacc , we consider two situations. First, an EV makes no Expression (1), in which c represents 1/3600/1000 J/kW h, gains from kinetic energy through regenerative braking when and t denotes a time. slowing the vehicle. Second, we chose ⌘ = 0.7 empirically 1 for estimating the worst conversion efficiency. et = c(( ⇢Cd Av 2 + µM g cos ↵ + M g sin ↵ 2 dv 1 4. EXPERIMENT +(M + Mi ) ) ⇥ ⇥ v) [kW h] (1) dt ⌘ This section presents comparison of Emin and Emax with Ereal . The actual energy consumption was Ereal for our Finally, Emin is calculated as the summation of et (Ex- experiment. We used trips of two types for experimenta- pression (2)). Also, n represents the number of simulation tion: long trips and short trips (Table 2). Figure 1 presents logs of an OD trip. the altitude and distance of experiment trips. We ignored n X charging spots on short trips because the trip distance is Emin = et [kW h] (2) sufficiently short that additional charging is not required. t=0 4.1 Long trips 3.2 Emax Calculation This subsection presents our description of how to calcu- 4.1.1 Experiment conditions late maximum energy Emax . We define Emax as “energy We conducted EV driving experiments for long trips in consumption when an EV runs at constant speed vc , and Hokkaido in 2017 and 2018. Hokkaido has an area that Figure 2: Experiment Routes in Hokkaido in 2017. Figure 4: Ereal in Outward Trips. Figure 3: Experiment Routes in Hokkaido in 2018. is the longest interval in Japan between charging stations. In this area, a fundamental problem arises of whether an EV can arrive at the next charging station after it leaves a charging station. Therefore, we conducted experiments on three routes to ascertain which route is best for an EV driver. We simulated kitami 1 and kitami 2 using the system in Figure 5: Ereal in Homeward Trips. 2017. Figure 2 shows kitami 1 and kitami 2 routes. We designated charging points as CPs. The kitami 1 travel 4.2.1 Experiment conditions distance is greater, but the elevation di↵erence is smaller. After we accumulated EV energy consumption logs[10] in Furthermore, the kitami 2 travel distance is shorter, but a database, we evaluated the accuracy of Emin and Emax the elevation di↵erence is greater. We simulated kitami 10 using EV energy consumption logs accumulated from daily in 2018 (Fig. 3) because a new charging station is located commuting. We therefore accumulated a large amount of there. one commuter’s data. The total number of trips was 786: outbound trips were 434, homeward trips were 352. There- 4.1.2 Experiment results fore, we use these logs as Ereal . We then compare Emin and Table 3 shows the pre-estimated results. We conducted Emax with Ereal to evaluate their accuracy. EV driving experiments using a Nissan Leaf (Nissan Motor Co. Ltd.) as the experiment EV in 2017 and 2018. We used 4.2.2 Experiment results estimated energy consumption logs[10] calculated from GPS Table 5 presents pre-estimated results. Figures 4 and 5 data as Ereal . In 2017, the experiment EV’s remaining bat- portray histograms of Ereal values. These graphs show that tery charge was the equivalent of 13.4 kWh when its battery any Ereal values are always between Emin and Emax . was 80% charged. That charge was achieved through charg- ing time of 30 min. We selected and ran kitami 2 because 4.3 Overall experiment results kitami 1 Emin was 15.27 kWh (greater than the 13.4 kWh As shown in Figure 6, we verified that any Ereal values are @80%), as Table 3 shows. These calculations indicate that always between Emin and Emax , even though EVs run on the EV can run the whole kitami 10 route if the remaining long trips or short trips. Ereal of outward and homeward battery charge is greater than 15.01 kWh. The experiment are mean values of numerous accumulated trips. EV was 16.7 kWh @100% (more than 15.01 kWh). There- fore, we chose the kitami 10 route for the 2018 experiment. Table 4 shows Ereal for the actual driving experiments. We 5. CONCLUSION can infer that the system outputs “Emin and Emax ” are As described in this report, we proposed a system for pre- correct because Ereal is between Emin and Emax . estimation of maximum and minimum electric vehicle (EV) energy consumption for use with unfamiliar roads. We de- 4.2 Short trips fined the minimum energy consumption is achieved when an Table 3: Emax and Emin in Long Trips Origin Origin CP CP CP CP Route ID to CP to CP to next CP to next CP to Destination to Destination Emax Emin Emax Emin Emax Emin kitami 1 19.17 kWh 15.27 kWh - - 6.62 kWh 5.88 kWh kitami 2 12.20 kWh 9.18 kWh - - 5.95 kWh 5.75 kWh kitami 10 15.01 kWh 11.98 kWh 4.16 kWh 3.29 kWh 6.62 kWh 5.88 kWh on Intelligent Transportation Systems (ITSC), pages 2319–2324, 2016. [2] M. Eshani, Y. Gao, S. Gay, and A. Emadi. Modern electric, hybrid electric and fuel cell vehicles 2nd.edition. In Power electronics and applications series, CRC press, 2010. [3] Y. Fei, W. Guoyuan, K. Boriboonsomsin, and M. J. Barth. A hybrid approach to estimating electric vehicle energy consumption for ecodriving applications. In 2016 IEEE 19th International Conference on Intelligent Transportation Systems (ITSC), pages 719–724, 2016. [4] J. Felipe, J. C. Amarillo, J. E. Naranjo, F. Serradilla, and A. Daz. Energy consumption estimation in electric vehicles considering driving style. In 2015 IEEE 18th International Conference on Intelligent Transportation Systems, pages 101–106, 2015. Figure 6: Overall Experiment Results. [5] S. Grubwinkler, M. Hirschvogel, and M. Lienkamp. Driver- and situation-specific impact factors for the Table 4: Ereal on Long Trips energy prediction of evs based on crowd-sourced speed Origin CP CP profiles. In 2014 IEEE Intelligent Vehicles Symposium Route ID to CP to Next CP to Destination Proceedings, pages 1069–1076, 2014. Ereal Ereal Ereal [6] M. Ito, T. Shimoda, and K. Maema. Prediction kitami 2 9.66 kWh - 5.81 kWh method of cruising range using probe data for electric kitami 10 13.04 kWh 4.11 kWh 6.40 kWh vehicle. In 20th ITS World CongressITS Japan, pages 1–10, 2013. Table 5: Emax and Emin in Short Trips [7] D. Karbowski, V. Sokolov, and A. Rousseau. Vehicle Route ID Emax Emin energy management optimisation through digital outward 5.00 kWh 3.10 kWh maps and connectivity. In 22th ITS World homeward 4.08 kWh 2.83 kWh CongressITS Japan, pages 1–10, 2015. [8] C. Kurtulus and G. Inalhan. Model based route guidance for hybrid and electric vehicles. In 2015 EV travels at a constant speed. Maximum energy consump- IEEE 18th International Conference on Intelligent tion occurs when an EV travels at a constant speed, but Transportation Systems, pages 1723–1728, 2015. also stops at controlled intersections and set places, such as [9] A. Styler, A. Sauer, I. Nourbakhsh, and sightseeing spots and stores. H. Rottengruber. Learned optimal control of a range Moreover, we conducted actual driving experiments, which extender in a series hybrid vehicle. In 2015 IEEE 18th yielded actual energy consumption logs with data in a range International Conference on Intelligent Transportation showing the estimated minimum energy consumption and Systems, pages 2612–2618, 2015. estimated maximum energy consumption. Our next chal- [10] T. Uemura, Y. Kashiwabara, D. Kawanuma, and lenge is to estimate a range that is specialized for individuals T. Tomii. Accuracy evaluation by gps data correction and routes using numerous daily life logs. Another challenge for the ev energy consumption database. In Adjunct is consideration of an air conditioner’s energy consumption Proceedings of the 13th International Conference on to output more correct EV energy consumption. Mobile and Ubiquitous Systems: Computing Networking and Services, pages 213–218, 2016. 6. ACKNOWLEDGMENTS [11] Q. Yang, K. Boriboonsomsin, and M. Barth. Arterial Part of this work is supported by JSPS KAKENHI Grant roadway energy/emissions estimation using Number 18K11750. modal-based trajectory reconstruction. In 2011 14th International IEEE Conference on Intelligent Transportation Systems (ITSC), pages 809–814, 2011. 7. REFERENCES [12] Y. Zhang, W. Wang, Y. Kobayashi, and K. Shirai. [1] M. De Souza, M. Ritt, and A. L. C. Bazzan. A Remaining driving range estimation of electric vehicle. bi-objective method of traffic assignment for electric In 2012 IEEE International Electric Vehicle vehicles. In 2016 IEEE 19th International Conference Conference, pages 1–7, 2012.