2.1. PROBLEM FORMULATION 7
2.1.3 DRIVING EVENT
A driving event is a driving maneuver, such as acceleration, deceleration, turning, and lane
change, which can be used to identify driving styles [28]. As mentioned previously, this study
mainly focuses on longitudinal motion control, hence the adopted driving events are defined
as [29] follows.
(1) Event 1: 0–50 km/h acceleration. In this event, the car is accelerated from 0–50 km/h.
e vehicle acceleration, jerk, and the time taken in this process are typical performance indices.
is event is used to optimize and evaluate the dynamic performance and ride comfort under
different driving styles.
(2) Event 2: 50–0 km/h deceleration. In this event, the car is decelerated from 50 km/h to 0.
e deceleration and the time taken in this process are typical performance indices. e energy
recovered during the braking process can be used to evaluate energy efficiency. is event is
used to optimize and check vehicle’s dynamic performance and energy efficiency under different
driving styles.
(3) Event 3: driving cycle. Although the energy consumption of the vehicle can be eval-
uated in the above two events, the time duration of an acceleration or deceleration procedure
is relatively short, making it difficult to evaluate energy consumption at the vehicle level. us,
the ECE driving cycle is adopted for measuring energy efficiency under different driving styles.
e ECE driving cycle, which is a series of data points representing the vehicle speed vs. time,
exhibits the typical driving conditions of a car in urban areas [17]. It is usually adopted to carry
out road testing for studying the fuel economy of a passenger car.
2.1.4 DRIVING STYLE RECOGNITION
To identify driving style for control synthesis and system optimization, a driving style recogni-
tion (DSR) algorithm is developed using unsupervised machine learning with partially labeled
data. e data set is collected in the road tests with a Sedan-Type vehicle, and it is comprised
of 9 real life cycles covering over 500 km. e data can be overall classified into three groups
according to the driver feedback as aggressive, conservative, and moderate. ese three driving
styles are firstly defined as [29–34] as follows.
(1) Aggressive: Aggressive drivers exhibit frequent changes in throttle and brake pedal
positions [32]. ey drive with sharp and abrupt accelerations and decelerations, aiming at ve-
hicle dynamic performance. is kind of behavior would result in higher fuel consumption and
increased likelihood of accidents [29].
(2) Conservative: Conservative drivers often exhibit mild operational behaviors with small
amplitudes and low-frequency actions on a steering wheel, accelerator, and brake pedal [33].
ey value energy efficiency and ride comfort, and avoid abrupt variations of vehicle state.
8 2. CO-DESIGN OPTIMIZATION FOR CYBER-PHYSICAL VEHICLE SYSTEM
150
100
50
0
500 1000 1500 2000 2500 3000
City
Rural
Highway
Road Type
t (s)
Vehicle Speed (km/h)
Figure 2.3: e real life route used for DSR experimental validation.
(3) Moderate: Moderate drivers are positioned between the above two. ey would like to
balance multiple performances, such as vehicle dynamic performance, ride comfort, and energy
efficiency [29].
e unlabeled data set is pre-processed for driving events detection and statistics extrac-
tion. A total amount of six signals is used: throttle pedal position, brake light switch, longitudi-
nal and lateral accelerations, steering wheel angle and vehicle speed. Five statistics are extracted
per event: maximum, minimum, mean, standard deviation, and root mean square. e reduced
set of signals is clustered using Gaussian Mixture Models (GMM), which generates the DSR
classification algorithm to be implemented onboard. e performance of the DSR algorithm
is validated against the subjective labels and further tested with a new set of data from a new
real-life route with changeable road type, as shown in Fig. 2.3. is new data set is collected by
a Sport Utility Vehicle (SUV)-type vehicle with a different driver.
Table 2.1 shows the results of the SUV driving data using the developed DSR algorithm.
So as to quantitatively evaluate the performance of the algorithm, the driving cycles are clas-
sified per events using the aggressiveness index. e aggressiveness index is transformed from
the classification into an equivalent index, assigning an increasing value from 0–1 to the dif-
ferent events based on the level of aggressiveness [34]. To provide further information about
the robustness of each classification, the number of events identified is included in brackets and
italics. According to the results, the conservative cycle is classified as the least aggressive one,
particularly by acceleration and brake events analysis. e moderate cycle is situated between
the aggressive and conservative ones. While the aggressive cycle is identified as the sportiest
one, but it has a similar braking level with the moderate one, agreeing with driver’s feedback.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.