A driving cycle construction methodology combining k-means clustering and Markov model for urban mixed roads

Abstract

With the rapid urban expansion in China, there is considerable development of road networks, mainly comprising arterial roads, expressways, and loop-lines in urban regions. In addition to typical short trips, certain number of long trips are also included in the driving behavior for urban areas. In view of the new characteristics of urban mixed roads, a driving cycle construction methodology combining k-means clustering and Markov model method is proposed based on about 2.4 million seconds of driving data collected from seven passenger cars without route planning. In order to cluster micro trips into appropriate categories corresponding to different types of roads, the Silhouette equation is introduced to determine the appropriate clustering strategy and the error of artificial factors is restrained. The constructed driving cycle, comprising a 1200 s speed-time series, is developed based on the assessment of speed–acceleration frequency distribution and estimation of six characteristic parameters. According to the results, the average difference rate of six characteristic parameters is only 4.31%, which demonstrates the effectiveness of the proposed method.

Comparing Driving Cycle Development Methods Based on Markov Chains

“…The most widely used definition of a microtrip is the sequence between two successive stops ( 7 ). That definition was used among others in recent studies conducted by ( 9 , 10 ) and ( 11 ). However, this segmentation method is not appropriate for urban areas where traffic and urban roads cause stop-and-go driving behaviors, which create very short and biased microtrips ( 8 , 12 ).…”

“…This method involves a simple and time-efficient distribution of the microtrips according to their average speed in predefined speed intervals. Other studies preferred classifying the microtrips using clustering algorithms as k-means ( 11, 14–16 ) and k-medoids ( 17 ), combined with different distance measurements.…”

“…In other studies, a principal component analysis (PCA) was conducted to consider more characteristics of the microtrips. Yuhui et al ( 15 ) and Peng et al ( 11 ) characterized the microtrips with approximately 15 parameters linked to speed, acceleration, and driving mode proportions. They applied the PCA on the 15 characteristic parameters and measured the distance between microtrips in their three or four new dimensions.…”

“…The microtrips are then concatenated using the transition matrix until the desired cycle length is obtained. The driving cycle length normally varies between 20 min ( 11 , 13 , 15 ) and 46 min ( 20 ). Comparison of the microtrip and Markov-chains methods have more than once concluded that the Markov-chains method better represents the modal events of the database, and thus better represents driving behaviors ( 18 , 19 ).…”

Improved K-medoids algorithm-based clustering analysis for handle driving force in automotive manual sliding door closing process

“…Clustering results based on the curve shapes of the handle driving force could provide insights of the subjects’ behaviors during closing the sliding door. Although clustering algorithms have been widely used in automotive engineering, 9,10 nobody has implemented clustering methods to handle driving force problem yet.…”