Efficiency for Large Fleets: An Analysis of How Industrial Customers Use Electric and Hybrid Vehicles

This paper presents methods and results which allow an analysis of relevant driving parameters of hybrid and electric vehicles. In order to gain crucial insights into how industrial customers use hybrid and electric vehicles, the paper investigates the following parameters: information about odometer, charging processes and battery charging levels. The data used for this purpose was provided by the Canadian fleet management company Geotab Inc. They were evaluated by means of ‘Google BigQuery’ and the statistics programme ‘IBM SPSS Statistics’. It turned out that correlations between ‘charging time’ and ‘battery charging level’ exist, as well as between ‘battery level’ and ‘distance per day’. One of the main questions in the present study asks whether long charging times of car batteries lead to decreased average battery charging levels. As a result of this study, the longer a hybrid vehicle is charged per day, the lower sink its average battery charging level. The findings of this research help managers of car fleets to enhance their existing fleet management for establishing more efficient fleets with respect to ecological and economical aspects. Our research is particularly significant as this will save money across such fleets worldwide, and at the same time, preserve the environment as much as possible.


Introduction
Due to recent considerations of how to create options for more sustainable ways of life, vehicles powered by alternative driving systems play an increasingly important role in our society (Kampker et al. 2018, pp. 143-144). The acceptance of a such new technology by customers is one of most important aspects regarding success. The more knowledge about customer behavior is collected and processed to understand their needs, the better these needs can be used to enhance these technologies. (Manowicz 2018, p. 59) The automotive industry therefore has multiplied efforts to develop and produce significantly higher numbers of electric and hybrid vehicles. It has been projected, for instance, that by 2030, at least 6 million electric vehicles should be circulating on German roads (Propfe et al. 2013, pp. 6-7). Many companies see the changeover to electric vehicles not only as an important asset for the environment, but also as an opportunity to enhance their public image (Globisch et al. 2018, pp. 10-11). Wikström et al. (2015, p. 2) state that it is possible to run large fleets of plug-in electric vehicles (PEVs). Consequently, the air quality improves in cities where these fleets are mostly deployed. Further, 'the utilisation degree of vehicles operating in commercial fleets is relatively high and therefore it could therefore [sic] be

Methodology
The starting point of the analysis was to go carefully through the dataset offered by the company Geotab in the format of csv-files and to select the most relevant categories for the present investigation. The dataset consists of more than 100.000.000 entries, which are divided by 14 different parameters. For the present study, relevant entries were selected by employing the powerful data warehouse Google BigQuery. Downsizing the dataset was done in three steps. In the first step, only vehicles sending data about their odometers were taken into account. A further step reduced the sample to include vehicles with data recording their charging processes. Finally, step three reduced this sample of vehicles to those for which details about their odometers, their charging processes and their battery charging level were provided. Following the creation of our data base, these entries were transferred to Microsoft Excel for evaluation.
In order to be able to evaluate the data statistically, certain values relevant to the investigation had to be calculated. The newly calculated values are the driven distance, the number of charging processes, the average charging time, and the average battery charging level. These values were tested with respect to statistical significance by running these through the programme 'IBM SPSS Statistics'. This programme was chosen as it is particularly well adapted to visualize statistical distributions and connections and to carry out additional statistical calculations.

Database
In the statistical analysis, average values of the above-mentioned parameters as well as correlations between different parameters were examined. In other words, only vehicles were included in the analysis for which the entries provide information about odometer, charging processes and battery charging levels. An additional criterion was applied: only vehicles that were used for exactly 365 days were taken into account, the start date for all of these being the 1st of September 2017.  Vol.10, No.3, 2019 Some vehicles displayed an extraordinarily long 'charging time'. As this would impinge the analysis unevenly, the research team decided to treat these as outliers and therefore, to place an upper limit of 1836 minutes to 'charging time' for all vehicles used in the analysis. This corresponds to a charging time of exactly 30.6 hours, the reference value of the maximum charging time of a Tesla Model S on a standard 230V household socket with 16A. According to the prerequisites described above, the following vehicles remain in the data set below (Table 1)

Parameters
The statistical analysis of these 88 hybrid vehicles examined the parameters 'distance per day', 'charging time' and 'battery charging level'. To calculate the parameters for each selected hybrid vehicle, the following definitions were developed: The 'entire distance' denotes the sum of all distances (in km) travelled by a vehicle during the specified 365days period. The 'entire charging time' sums up the duration of each charging process per vehicle in the 365-day time interval. The 'battery charging level' is defined as the arithmetic means of all battery charging levels of the respective vehicle stored in the dataset over a period of 365 days.

Statistical Distributions of Parameters
The statistical distributions of the selected parameters (described in the methodology section) were, in a first step, graphically determined. Additionally, the arithmetic means of the parameters 'distance per day', 'charging time' and 'battery charging level' were ascertained. The following chart shows the statistical distributions as well as the respective averages of the parameters described above for all examined 88 hybrid vehicles.  The analysis shows the following statistical distribution of the 'distance per day' for all 858 vehicles.

Test on Normal Distribution
This test is calibrated to check the statistical distribution of the parameters 'distance per day', 'charging time' and 'battery charging level' with respect to normal distribution (for details on the method, see Bortz et al. 2010, pp. 70-74). The procedure was conducted along the lines of Shapiro-Wilk (for details on the method, see Razali et al. 2011, pp. 21-33). The result -as to whether normal distribution does or does not exist -decided on which procedure was used in the subsequent correlation analysis. In order to be able to decide whether the parameters were to be tested for correlation according to parametric or non-parametric test methods, a test for normal distribution of the underlying data had to be carried out. The data of the parameters 'distance per day' and 'battery charging level' for the 88 commercially used hybrid vehicle models Volt (Chevrolet) and Outlander (Mitsubishi) are distributed normally, given the statistical significance (for details on the method, see Bortz et al. 2010, pp. 100-102) of 0.080 > 0.05 and 0.520 > 0.05. The data of the parameter 'charging time' are not distributed normally due to a statistical significance of 0.00 < 0.05.

Testing Correlation between Parameters
In the next step, the interrelations and correlations between the parameters 'distance per day', 'charging time' and 'battery charging level' were examined. In the case of normal distribution of data, correlations were analysed according to Pearson, and in the opposite case, according to Spearman. The parameters 'distance per day' and 'charging time' hardly correlate with each other: they do so only with a correlation coefficient of -0.062, calculated according to Spearman. Given the statistical significance of 0.568 > 0.050, the null hypothesis 'no correlation' has to be assumed. Thus, the correlation is not significant.

Comparison of the Vehicle Models
Finally, the two vehicle models Volt (Chevrolet) and Outlander (Mitsubishi) were compared with each other. The main question was whether the two models show significant differences with respect to certain parameters. The comparison came to the following results: the mean values of the parameter 'distance per day', 'charging processes' and 'charging time' of the two models differed. For this investigation, the parameter 'charging processes' was defined as including all charging processes during the time 365-day period.  Vol.10, No.3, 2019 In order to verify whether the differences between the arithmetic means of the parameters 'distance per day', 'charging processes' and 'charging time' are statistically significant, the U-test according to Mann-Whitney (for details, see Bortz et al. 2010, pp. 130-133) was applied. Its implementation arrives at the following results. Given the statistical significance values of 0.010 < 0.050 and 0.001 < 0.050, the null hypothesis 'there is no significant difference' for the parameters 'distance per day' and 'charging time' has to be rejected. This means that the differences between the average values of the parameter 'distance per day' and 'charging time' between the vehicle models Volt (Chevrolet) and Outlander (Mitsubishi) are statistically significant. The Outlander (Mitsubishi) is moved 22.8 km more on verage per day than the Volt (Chevrolet). In addition, the Volt (Chevrolet) charged about four times longer on average than the Outlander (Mitsubishi). Given the statistical significance of 0.447 > 0.050, the null hypothesis 'there is no significant difference' has to be assumed for the parameter 'charging processes'. This means that the difference of the arithmetic mean of the parameter 'charging processes' between the vehicle models Volt (Chevrolet) and Outlander (Mitsubishi) is not statistically significant.

Limitations
Overall, the entries of the Geotab dataset provided odometric values for 858 electric and hybrid vehicles which were used over the time period of 365 days. However, values for 'charging processes' and 'battery charging time' were provided only for 90 models, including Ampera (Opel), Soul EV (KIA), Volt (Chevrolet) and Outlander (Mitsubishi). Therefore, it was not possible to include a larger sample of vehicle models in the analysis with respect to 'charging processes' and 'battery charging time'.

Conclusions
Increasing efficiency in managing fleets of electric and hybrid vehicles is a crucially important economic aspect for industrial companies and operators. The present paper investigated whether significant interrelations and correlations exist between the parameters 'distance per day', 'charging time' and 'battery charging level' for such vehicles. For this study, large datasets had to be managed and analysed graphically. This was possible through using 'Google BigQuery' and 'IBM SPSS Statistics'. The results of the analysis highlight the characteristics of electric and hybrid vehicles used 365 days in the time period between the 1st of September 2017 and the 31st of August 2018. These vehicles are moved 103.5 km per day on average. The average 'charging time' of vehicle models Volt (Chevrolet) and Outlander (Mitsubishi) is 89.1 minutes per day. Their batteries are charged at 48.3% on average. In addition, a significant correlation exists between the average 'charging time' per day and the average 'battery charging level'. The longer a hybrid vehicle is charged per day, the lower sink its average 'battery charging level'. Additionally, higher average 'battery charging levels' allows driving longer distances per day. Based on the results it is necessary that managers of fleets control the 'charging time' of their commercially used electric and hybrid vehicles closely.
In sum, the analysis of how industrial customers and fleet operators use their electric and hybrid vehicles provides significant insights. It also allows developing new solutions and strategies to increase the efficiency of managing large fleets of industrial customers. This scientific investigation acts as a spring board for further research in the field of increasing the efficiency in managing large fleets, both in terms of economic and ecological benefits. There are various possibilities in which directions future research could go. For instance, it would be fascinating to include improvements of the technology of cars, such as develop batteries that have a much longer life, and thus be more environmentally friendly in the production process. This would enable customers to plan their itineraries with more flexibility as batteries will not have to be charged as frequently; this in turn would again increase the efficiency for fleet managers. There is an array of further research options, however, setting all of them out requires another research paper.