Binary Recursive Partitioning Method for Modeling Hot-Stabilized Emissions From Motor Vehicles

Publisher Website
Google Scholar
Abstract
An alternative statistical modeling approach, hierarchical tree-based regression (HTBR), is presented for developing modal correction factors for hydrocarbon (HC) emissions from motor vehicles. The term modal refers to operating modes of vehicle activity such as cruise, idle, deceleration, and acceleration. Explanation of the statistical theory is provided, followed by a presentation of specific modeling results for HCs. The modeling results are based on 4,800 vehicle emissions tests representing 29 laboratory testing cycles. HTBR methods are indicated to overcome statistical difficulties that are problematic for classical ordinary least-squares (OLS) regression, a commonly applied statistical technique for analyzing emissions data. HTBR methods are more adept at treating interactions and monotonic transformations on independent variables, better at handling categorical independent variables with more than two levels, not adversely affected by multicollinearity, and good at capturing nonadditive behavior across the range of independent variables. Unfortunately, HTBR theory is less well developed than OLS regression theory, and statistical parameter properties, such as efficiency, unbiasedness, and consistency, need further development. The HTBR modeling results for HCs are insightful. Hydrocarbon emissions from normal-emitting motor vehicles are most sensitive to changes in power (instantaneous speed2 ƃ acceleration) requirements of a given driving sequence, while high-emitting vehicles are sensitive to both the amount of idle activity and positive kinetic energy (instantaneous speed ƃ acceleration) in a given driving sequence. Vehicle model year, engine size (cubic centimeters of displacement), curbside weight, and fuel delivery type (fuel injected, throttle body injected, carbureted), also were indicated to influence emission rates. Finally, high- and normal-emitting vehicles are sensitive to different operational and vehicle specific factors.

This publication has 2 references indexed in Scilit: