Development of real-world Driving Cycles for Battery Electric Vehicles

Summary Driving cycles are used in various aspects of vehicle development and for the homologation of new vehicles. The character and duration of a cycle have a huge impact on the energy consumption for longitudinal dynamics and comfort functions. Current driving cycles are widely criticised for their insufficient representation of real-world driving, especially for battery electric vehicles (BEV). Therefore we developed three driving cycles for BEV that are based on empiric data from a large-scale field operational test with electric vehicles (project CROME) for future application in research and policy making.


Introduction
Driving cycles are an important instrument in the development of new vehicles and for the evaluation of vehicle characteristics like energy consumption and emissions. In recent years, the discussion about the appropriate cycle has gained momentum in science [1], [2] and in the public domain [3] as the gap between real-world consumption and driving cycle consumption increased due to optimized adaption to the certification process. In [4] it is shown that maximum and average propulsion power differ for three vehicle classes between Urban Dynanometer Cycle and US06 Cycle by more than factor 2. For battery electric vehicles (BEV), the difference in energy consumption can reach more than factor 2 and even more than factor 5 in case of the maximum driving torque depending on the cycle [5]. This demonstrates that the choice of the appropriate driving cycle is absolutely crucial for a meaningful evaluation of vehicles. For that reason we decided to build up empirically based real-world driving cycles for electric vehicles for future application in research and policy making. This work is presented here an excerpt from the corresponding author's doctoral thesis [6].

Empiric foundation
The data, on which the driving cycles are built, has been gathered in the project CROME -Cross-border Mobility for Electric Vehicles [7], [8]. In this field operational test more than 100 BEV of 6 types have been employed in corporate fleets in France and Germany by average users. Almost 3 years of data collection in 80 vehicles have led to more than 125,000 trip records with a total length of about 600,000 km in binned data or continuous curve signals. The driving cycles presented here are founded on 16,690 trips, which consist of a total number of 88,431 driving sequences. They cover a driving distance of ca. 129.000 km and have been recorded with continuous curve signals at an acquisition rate of at least 1 Hz. Overall the data shows a ratio of 74 % urban trips and 26 % rural trips. Rural trips are defined as trips that exceed 70 km/h at least once and show a speed profile of at least 20 % of the driving time beyond 60 km/h. The data shows a focus on rather short trips with a mean value of 8.4 km and a median of 5.2 km. This results mainly from the locally focused mobility of the users, but -in parts -as well from the availability of alternative vehicles with combustion for longer trips. The BEV have in many cases been used as part of heterogeneous fleets in combination with other vehicles.

Methodology
The goal of the development of the driving cycles was to gain the best possible representation of the data basis in one representative trip per usage domain. The three domains, which have been defined as targets, are predominantly urban driving, predominantly rural driving and the mixed overall operation with a ratio of urban to rural segments as it is found in the data basis.
The first step to build up each driving cycle is a statistical analysis to define target vectors for the total length, the duration, the proportion of urban to rural driving and the number of driving sequences to gain the best possible representation of the driving data. The following procedure is based on the methodology that has been used in [9]. Hereby 13 further statistical parameters are derived from the data base to complete the target values for the driving cycle. The parameters are based on the evaluation of [9], [10] and [11]. The complete overview of the parameters employed is given in table 1. Average longitudinal deceleration m/s 2 5 Average duration of driving sequences s 6 Proportion idling % 7 Proportion cruising % 8 Proportion creeping % 9 Proportion acceleration % 10 Proportion deceleration % 11 Number of acceleration-deceleration changes 12 Root mean square longitudinal acceleration m/s 2 13 Normalised positive acceleration kinetic energy m/s 2 The borders of speed and lateral acceleration for stop, cruising, creeping, acceleration and deceleration can be found in table 2.
After the statistical analysis, all trips of the database are divided into micro-trips, what results here in a total number of 88,431. A micro-trip is defined as the phase with a positive driving speed between two stops as shown for example in Figure 1 with 5 micro-trips. Every micro-trips keeps linked with the stop phase that occurred before it to have realistic idle phases in the cycle. The micro-trip pool is in the next step cleaned to remove strong statistical outliers. Finally, random combinations of micro-trips with the statistically pre-defined total number of micro-trips are built-up and all parameters evaluated in comparison to the target vector. In total 100,000,000 combinations are evaluated and compared to find the driving cycle that represents the data best. This process is repeated for each of the three driving cycles with it specific target values.

Results
In total three driving cycles have been built up to represent the real-world usage of BEV in corporate fleets in central Europe.

Fleet-BEV-Cycle
The Fleet-BEV-Cycle (see Figure 2) for combined usage and the Fleet-BEV-Urban-Cycle and the Fleet-BEV-Rural-Cycle for urban and rural driving. The Fleet-BEV-Cycle shows an average deviation of the target values of 2.8 % and a maximum deviation of 8 %. It is intended to represent the overall usage of the BEV and therefore consists of four of urban as well as one rural micro-trip. As it is found in the data base, the driving cycle reflects predominant urban operation. In sum 13.7 % of the actual driving time without idle phases shows a speed of 60 km/h or more. The Fleet-BEV-Cycle has phases with a speed limit of 50 km/h and with 30 km/h as well as an initial sequence of 27 s with a top speed of 21 km/h as it occurs when a large parking lot has to be left. Key characteristics of the three cycles can be found in Table 3 and the corresponding speed values are found in Appendix A.
The longitudinal characteristic of the cycles can be visualised with the speed acceleration probability distribution (SAPD). The Fleet-BEV-Cycle has the core of the usage in classes of constant driving to soft acceleration at urban speed classes (see Figure 3). But there is also a significant share in the class between 70 km/h and 80 km/h resulting from the rural driving sequence. Accelerations and decelerations range to ca. +/-2.2 m/s 2 .

Fleet-BEV-Urban-Cycle
The Fleet-BEV-Urban Cycle ( Figure 4) represents a purely urban trip. In comparison to the Fleet-BEV-Cycle and the Fleet-BEV-Rural-Cycle it is shorter in distance and time, has a lower average speed and a higher proportion of idle-time. At the same time it consists of seven micro-trips, which is the highest number of the three cycles as urban micro-trips are usually shorter than rural micro-trips. The average deviation of the target values is 4.3 % and the maximum is 9.9 %.   Figure 6 shows the Fleet-BEV-Rural-Cycle which is intended to represent a typical rural trip with a fleet BEV. It matches the target values with an average deviation of 2.1 % and a maximum deviation of 5 %. Here again, the SAPD matrix is used to visualize the longitudinal acceleration per speed class (see Figure 6). The profile is now comparably narrow, resulting from a higher proportion of constant or nearly constant driving. The main proportion of driving happens beyond urban speed limits but the share of urban driving is still noteworthy. Rural driving shows decelerations around 2 m/s 2 at speed levels above 60 km/h and there is even one strong deceleration with in peak approx. 3 m/s 2 . Table 3 gives an overview of key characteristics of the three cycles.

Comparison to other Driving Cycles
The The comparison with the Worldwide harmonized Light duty driving Test Cycle (WLTC) [13] for class 3 vehicles reveals the biggest differences in the maximum speed (120 km/h), the distance (23 km) and the duration (30 min). This brings the WLTC far away from representative for the observed BEV usage as only 3.5 % of the observed trips are 23 km long or more. Especially for the representation of all processes that tend towards a thermo-dynamic equilibrium like the tyre temperature, that largely influences rolling resistance [14], or the climatisation of the interior, it is absolutely crucial to have realistic driving time, distance and speed profiles.
The volatility of the three driving cycles, as they are found in all real-world speed profiles might turn out as difficulty for the adaption of the cycle for bench tests. Therefore a series of test drives with trained drivers should be conducted to evaluate the deviation of driving speed in set-actual comparison. Maybe, for this purpose it will turn out necessary to smooth the speed profile. But that process should lead to a revaluation of energy consumption of the adapted cycle towards the original cycle in a simulation environment. The potential drawback of driveability influences only bench tests with human drivers but does not affect the usage in simulation models. Being combined from real-world micro trips that are connected with idle phases the general driveability of the cycle is guaranteed.

Conclusion
In this paper we presented three driving cycles for electric passenger cars that have been built up on a large empiric basis from real-world driving of BEV in corporate fleets. The comparison with NEDC and WLTP reveals major differences in distance, duration and longitudinal dynamics that account for unrealistic consumption and an insufficient representation of everyday usage in those cycles. Therefore we recommend the family of the here presented Fleet-BEV-Cycles for the future research on battery electric passenger cars. The usage on test benches with human drivers might need slight adaption for enhanced drivability. To determine this, we recommend a test campaign for the cycles with trained drivers. For the usage in simulation models, the cycles are suitable right away with the values found in Appendix A to C. They can be employed for the evaluation, development or enhancement of components, full vehicles or the integration of BEV in traffic or energy research models.

Authors
Matthias Pfriem has been working as research assistant at the Institute of Vehicle System Technology at KIT from 2010 to 2015 to prepare his doctoral thesis. During this time he has organised and conducted the scientific analysis of data from three field operational tests with hybrid and full electric vehicles. The main focus of his work were the analysis of the real-world usage of those vehicles with regard to user acceptance and as a basis for demand-based vehicle specifications as well as the associated development of new methods of data logging.