Accurate Remaining Range Estimation for Electric Vehicles Joonki Hong, Sangjun Park, Naehyuck Chang Dept. of Electrical Engineering KAIST joonki@cad4x.kaist.ac.kr
Outline Motivation: Remaining range estimation Related works Our framework Experiment Conclusion 2
Fully Charged Ranges of EV Electric vehicles are emerging with their great advantages, but EV drivers suffer from the short driving range EV have 5X shorter fully charged ranges than that of ICEV (Internal Combustion Engine Vehicles) ICEV weight ICEV [kg] EV EV weight [kg] 2,400 Tesla model S Curb weight [kg] (kg) 1,800 1,200 600 Volt Soul Leaf Tesla Soul Cube Spark Twingo Toyota RAV4 0 Twizy 0 275 550 825 1,100 1100 Fully Fully charged range range [km] (km) 3
Range Anxiety Range anxiety is the fear that a vehicle may not have sufficient energy to reach its destination The range anxiety comes from the uncertainty of the remaining range Modern fuel gauges show the remaining range but it s only based on the driving history (past fuel consumption) 4
Range Anxiety in EV Most of the electric vehicle drivers always suffer from the range anxiety Limited EV charging facilities, long charging time Running out of battery while driving gives the same inconvenience as the vehicle breakdown 5
Range Anxiety in EV The statistics says that most of the EV drivers attempt only 70% of the estimated driving range with confidence The range anxiety in EV make the effective driving range of the EV even shorter Efforts to mitigate range anxiety More EV charging facility, higher density batteries, fast charging, and range extender Above solutions are time consuming or high cost Rav4 EV 112 km 160 km Model S 300 km 420 km 6 Twingo 1000 km
Paper Contribution In this paper, we provide accurate range estimation It mitigates the uncertainty of the driving range alleviates the range anxiety It restores the reserved range and extend the effective driving range Same effect of extended range without increasing the battery capacity Case 1 Case 2 7
Outline Motivation Related works History based RR Model based RR Our framework Experiment Conclusion 8
RR Estimation Framework The general remaining range framework can be simplified as following figure Is it at the destination? Driving start Input: route information, road slope and driving pattern Check the traffic, current velocity, weather, etc. Is the current SoC the lower limit? Step 1: Predict the future velocity profile with the route information, driving pattern, and speed limit Step 2: Estimate the remaining range with the EV power consumption model Driving end Print out the remaining range and the SoC 9
History Based RR History based remaining range estimation assumes the future power consumption is the same as the past power consumption Is it at the destination? Driving start Input: route information, road slope and driving pattern Check the traffic, current velocity, weather, etc. Is the current SoC the lower limit? Step 1: Predict the future velocity profile with the route information, driving pattern, and speed limit Step 2: Estimate the remaining range with the EV power consumption model Driving end p ˆf uture (x) p past (x) Print out the remaining range and the SoC Some works used regression based algorithm to improve the estimation accuracy p ˆf uture (x) = y (x) p Long 10
Model Based RR Model based remaining range estimation is naturally more accurate than the history based estimation There are some works about the model based range estimation They focused on predicting the future driving profile, but left the power model simplified SoC: 10% History based RR: Can t Reach destination Model based RR: Can reach destination 11
Paper Contribution In this paper, we focus on an accurate EV power model to achieve an accurate remaining range estimation Is it at the destination? Driving start Input: route information, road slope and driving pattern Check the traffic, current velocity, weather, etc. Is the current SoC the lower limit? Step 1: Predict the future velocity profile with the route information, driving pattern, and speed limit Step 2: Estimate the remaining range with the EV power consumption model Driving end Print out the remaining range and the SoC 12
Outline Motivation Related works Our framework Remaining range estimation EV power model Experiment Conclusion 13
Proposed Range Estimation We start with a basic assumption that the future driving profile is given EV Power Model Dynamics Motor Harvesting Empirical Future driving profile Battery model Remaining Range
Vehicle Dynamics Model Widely used vehicle power consumption model P dynamics = F ds dt = Fv =(F R + F A + F G + F I + F B )v (F R + F G + F I )v ( + sin + a)mv, F A F I F R (rolling resistance) C rr W F G (gravitational resistance) W sin F G F R F I (inertial resistance) ma F A (aerodynamic resistance) 1 2 C dav 2 15
Proposed Advanced Dynamics Model Motor efficiency actually differs dynamically according to the operating status = P P + k i! + k w! 3 + k c Q 2 + C F i (Iron and friction loss) k i! F w (Windage loss) k w! 3 F c (Copper loss) k c Q 2 Constant loss C T = P/v =( + sin + a)m P advanced = Tv+ C 0 + C 1 v + C 2 T 2 MAGSOFT corporation 16
Proposed Hybrid Power Model Advanced dynamics model ignores the drivetrain and ancillary losses in the estimation A pilot experiment to verify the adequacy of the advanced dynamics model Finally, we propose the hybrid power model including the quadratic term from empirical data T = P/v =( + sin + a)m 00 Velocity (m/s) P hybrid = Tv+ C 0 + C 1 v + C 2 v 2 + C 3 T 2 Power consumption (W) (kw) Power Power consumption (W) (kw) 5000 5 4500 4000 4 3500 3000 33 2500 2500 2000 22 1500 1000 1 500 Measured data1 quadratic data 0 0 2 4 6 8 10 Velocity Velocity (m/s) (m/s) Velocity (m/s) 10 0 2 Velocity 4 (m/s) 6 8 10 17
Regenerative braking model One of the most commonly used energy harvesting methods in the EV is regenerative braking The regenerative braking comes from the electromagnetic induction P induction /! Regenerative power can be modeled as follows P regen = v Battery Battery power Power (kw) (kw) 2 2 0 0-2 -2-4 = (460.53 J/m)v (333.92 J/s) Regeneration Cutback Cutback -4 0 2 4 6 8 10 0 2 4 6 8 10 Velocity (m/s) Velocity Velocity (m/s) 18
Power Models Vehicle dynamics model F A F I P vehicle =( + sin + a)mv Advanced dynamics model T = P/v =( + sin + a)m P advanced = Tv+ C 0 + C 1 v + C 2 T 2 Hybrid power model F G F R T = P/v =( + sin + a)m P hybrid = Tv+ C 0 + C 1 v + C 2 v 2 + C 3 T 2 19 Power Power consumption (W) (W) Power consumption (kw) 5000 5 4500 4000 4 3500 3000 3 2500 2500 2000 2 1500 1000 1 Measured data1 quadratic data 500 0 0 Velocity (m/s) 0 2 Velocity Velocity 4 (m/s) 6 8 10 (m/s)
Outline Motivation Related works Our framework Experiment Power modeling and validation Remaining range estimation Conclusion 20
Specification of the target vehicle We use a light-weight custom EV to verify the accuracy of power modeling and remaining range estimation Specification Curb weight: 481 kg Maximum velocity: 35 km/h 76.8 V, 48 Ah LiFePO4 battery pack Real-time monitoring: battery current, voltage, SoC, etc. SD card Status LCD: BMS, logging status, error code, etc. 21
Data logging We chose a regression based approach for the modeling For the model fidelity, we collect 6000s of driving data from various routes Slope (o) Slope (o) Slope degree (o) (o) degree (o) degree (o) 4 4 33 22 11 0 10 20 30 distance Distance (m) (m) 40 50 60 3 3 22 11 0 0 5 10 15 20 25 30 35 40 45 50 50 distance (m) Distance (m) 00-0.5-1 -1.5 0 20 40 60 80 100 120 distance (m) Distance (m) 22
Power consumption model Vehicle dynamics model T =( + sin + a)m, P dynamics = Tv = 0.59, = 12.63, = 1.46. Advanced vehicle dynamics model F A F I F G F R P advanced = Tv+ C 0 + C 1 v + C 2 T 2, = 0.33, = 10.70, = 1.09, C 0 =5.28,C 1 = 118.55,C 2 =0.0017. 23
Power consumption model Vehicle dynamics model T =( + sin + a)m, P dynamics = Tv = 0.59, = 12.63, = 1.46. Hybrid power model P hybrid = Tv+ C 0 + C 1 v + C 2 v 2 + C 3 Q 2, = 0.32, = 10.11, = 1.08, C 0 =5.28,C 1 =7.39,C 2 = 20.62,C 3 =0.0019. F A F I F G F R 24
Acceleration Model validation result Hybrid power model yield only 3.78 % error Power (kw) Power (kw) Slope (o) n Velocity (km/h) Degree (o) Acceleration (m/s 2 ) (m/s 2 ) Velocity (m/s) 10 55 00 55 00-5 -5 0 100 200 200 400 300 600 400 800 Time (s) 1000 500 1200 600 1400 700 1600 800 1800 900 2 Time (s) 0 200 400 600 800 Time (s) 1000 1200 1400 1600 1800 22 00-2 0 200 100 400 200 300 600 800 400 1000 1200 1400 1600 1800 Time (s) 500 600 700 800 900 Time (s) 40 Estimated PowerMeasured Power 0 100 200 400 200 300 600 400 800 Time (s) 1000 500 1200 600 1400 700 1600 800 1800 900 20 00 0 100 200 400 200 300 600 400 800 Time (s) 1000 500 1200 600 1400 700 1600 800 1800 900 25 Time (s) Time (s) Time (s) Measured Power
Test bench drives We perform 6 test bench drives for the remaining range estimation Each drives were performed in different driving manner Drive C Drives B, E Drive A Drive F Drive D 26
Remaining range estimation Measured Hybrid Advncaed Dynamics 60 Measured and estimated range (km) 7.5 Percentage error between measured and estimated range (%) 45 0 30-7.5-15 15-22.5 0-30 A B C D E F A B C D E F 27
Remaining range estimation Drives A and F Speed range: bellow 11.5 km/h Degree range from -0.6 to 0.9 7.5 0-7.5-15 -22.5-30 A B C D E F 28
Remaining range estimation Drives A and F 1.2 Hybrid Advanced Measured 1.2 Power (kw) 1 0.8 0.6 0.4 Power (kw) 1 0.8 0.6 0.4 0.2 715 720 725 730 735 740 745 750 Time (s) 0.2 715 720 725 730 735 740 745 750 Time (s) 12 12 Velocity (km/h) 11.5 11 10.5 10 Velocity (km/h) 11.5 11 10.5 10 9.5 715 720 725 730 735 740 745 750 Time (s) 29 9.5 715 720 725 730 735 740 745 750 Time (s)
Remaining range estimation Drives C and D Speed range: 5 km/h to 32 km/h Degree range: -0.6 to 0.85 7.5 0-7.5-15 -22.5-30 A B C D E F 30
Velocity (km/h) Power (kw) Remaining range estimation Drives C and D 10 8 6 4 2 0 Hybrid Dynamics Measured Power (kw) -2-2 -4 400 410 420 430 440 450 460 470 480 490 500-4 700 710 720 730 740 750 760 770 780 790 800 Time (s) Time (s) 30 30 25 20 15 10 5 Velocity (km/h) 5 400 410 420 430 440 450 460 470 480 490 500 700 710 720 730 740 750 760 770 780 790 800 Time (s) Time (s) 31 10 8 6 4 2 0 25 20 15 10
Remaining range estimation Closer look at drives C and D Hybrid Dynamics Measured Power (kw) 9 7 5 Power (kw) 9 7 5 3 400 410 420 430 440 450 Time (s) 3 400 410 420 430 440 450 Time (s) 9 9 Power (kw) 7 5 Power (kw) 7 5 3 3 700 710 730 720 740 750 700 710 730 720 740 750 Time (s) 32 Time (s)
Outline Motivation Related works Our framework Experiment Conclusions 33
Conclusions We achieve higher remaining range accuracy The absolute average errors for hybrid power model, advanced dynamics model, and vehicle dynamics model are 2.52%, 6.85%, 9.33% respectively The hybrid power model shows increased estimation accuracy not only in the total remaining range, but also in the instantaneous power estimation 34
Thanks for your attention!
Electric conversion 36
High speed custom EV 37