Development of representative driving cycles of the Tenerife metropolitan area through clustering methods

Abstract

In recent years, owing to the evolution of science and technology, more efficient methods of analysing high-dimensional data have been developed. Drawing on this progress, data science can be applied to the environmental sector, helping to determine more accurately the impact of vehicles emissions on the environment through representative driving cycles. This study aims to develop a methodology that helps to build representative driving cycles from a data set collected in the metropolitan area of Tenerife. The methodology proposed in this study consisted of the division of driving cycles into segments (Microtrips) and subsequently applying various clustering algorithms (k-means and Hierarchical clustering), following the application of a dimensionality reduction methodology (t-SNE and PCA). The results were split into groups with similar acceleration-related variables, representing the driving behaviours. . It was found that the highest quality clusters, assessed through silhouette coefficients, Calinski-harabasz and Davies Bouldin index, resulted from the utilization of a combination of t-SNE and k-means. The representative Microtrips were then merged to obtain the final cycle. This methodology seemed to be unable to satisfy the desired cycle duration without affecting the data's representativeness. However, when the final cycle was compared to the data set, the resulting discrepancies were deemed acceptable.