Urban Driving
3.4 Vehicle Specific Power
3.4.4 Categorizing VSP data
Instead of evaluating emissions data according to vehicle specific power (VSP) on a continuous, or second-by-second basis, binning or categorization methods have been developed in order to aggregate emissions data according to “VSP bins” based on
predetermined VSP ranges. The VSP binning approach was originally developed for the U.S. EPA by the North Carolina State University (EPA, 2002) and is the basis behind current emissions models such as MOVES. The method of categorizing vehicle emissions based on VSP bin is still universally used for on-road emissions research.
There are several advantages to using a grouping method for analyzing
continuous emissions data. Filtering data into groups will inherently mask some of the features within a dataset, however, despite this coarse application, VSP binning still maintains sufficient specificity to allow for meso-scale and macro-scale data analysis. If
implemented properly, modal analysis can also support a more detailed micro-scale analysis. Additionally, continuous on-road data possess a strong degree of
autocorrelation. Employing a modal or binning approach to classify the data eliminates the autocorrelation innately present throughout the dataset. Modal analysis is commonly used in the vehicle emissions arena, and its use here provides a basis for comparative purposes.
Two classification systems were developed for the EPA, and both were
implemented on the PHEV Sprinter data. A modal-based system specified four primary driving “modes” and established exact criteria for determining mode class for every second within a continuous, on-road dataset. The second classification scheme
segregated the emissions data into concrete VSP bins based on the instantaneous vehicle specific power calculated for every second of emission data collection.
VSP bins were originally created using vehicle-based variables such as vehicle class, mileage, age, engine size, vehicle weight, and technologies present on the vehicle.
Vehicle operation variables such as vehicle speed, acceleration, and vehicle specific power (used as a surrogate for power demand) were also included. Because of data availability, the researchers were also able to incorporate external parameters such as road grade, air condition usage, ambient temperature, and relative humidity into the inter-modal model. Hierarchal tree-based regression (HTBR), a forward step-wise variable selection method similar to forward stepwise regression in order to determine the end-points for each VSP bin (EPA, 2002), was used to divide the data into subsets, with each subset being more homogeneous than the amalgamation of the entire dataset. Each subset was required to be statistically unique from all other subsets, thus resulting in the
formation of discrete and unique VSP bins. HTBR was run in a partially supervised method, allowing the advantage of the inherent power in the statistical method’s design while making it somewhat impervious to the impact of artifacts and variability in the data that may or may not be important from a practical and intuitive basis (the supervised part).
Explicit criteria were established for defining model categories (bins). HBTR analysis showed that vehicle specific power was the dominant explanatory variable in predicting vehicle emissions, so the binning approach was based on VSP values. NCSU researchers required that each bin should have a statistically significant different average emission rate than any other bin and that no single bin should dominate the estimate of total emissions for a typical route or trip. Based on these criteria, no bin was allowed to explain more than, approximately, 10% of the total emissions.
VSP bins created for the EPA in the 2002 NCSU report were developed for light-duty gasoline vehicles. This binning procedure, producing 14 discrete VSP bins, was initially investigated for the PHEV study; however, the PHEV data did not give sufficient weight to all VSP bins, resulting in under representation of the higher bins. Because of this, the VSP bin model developed for heavy-duty vehicle classes, and used throughout the technical literature (Zhai et al., 2008 and Frey et al., 2007) was, instead, used
throughout the PHEV study. It was important that the binning approach adopted for this work fit the PHEV data to the best of its ability, primarily through sufficient data
representation of all bins in the binning technique. Table 3.6 defines the 8 VSP bins used throughout the PHEV analysis.
Table 3.6: VSP bins (Frey, 2007).
VSP Mode VSP Range (W/kg)
1 VSP ≤ 0
2 0 < VSP < 2
3 2 ≤ VSP < 4
4 4 ≤ VSP < 6
5 6 ≤ VSP < 8
6 8 ≤ VSP < 10
7 10 ≤ VSP < 13
8 VSP ≥ 13
The second data-categorization approach was based on four distinct driving modes: idle, acceleration, cruise, and deceleration (EPA, 2002). Explicit conditions were established for designating the operating mode of the vehicle. Periods of on-road
operation with zero velocity and zero acceleration were identified as idle mode. Cruising mode was defined as approximate steady-state operation with some allowable drift in velocity. In order for the vehicle to be designated as operating within acceleration and deceleration modes, several criteria were established to completely segregate acceleration and deceleration events from the oscillatory, transient operation that would naturally occur during cruise mode. First, in order for a vehicle to be designated as operating in acceleration mode, it must be identified as both moving and increasing in velocity, therefore both velocity and calculated acceleration must be greater than zero.
Additionally, one of two criterion must be met in order to accurately identify acceleration mode from natural, subtle velocity fluctuations that occur during cruising: first, the vehicle must be experiencing an acceleration rate of at least two mph/sec, or the vehicle mush have established a sustained, moderate acceleration rate of one mph/sec for a
was required for the vehicle to meet the demands of deceleration mode. Previous research has shown that the four designated driving modes result in statistically significant differences in emission rates (Frey et al., 2001, and Frey et al., 2002).
It was further found that the predictive ability of the modal model could be improved by employing VSP modal development within the operating modes cruise and acceleration (EPA, 2002). This additional step was not included in the PHEV analysis, however, since there has been no reference of this modal development for heavy-duty vehicles, only light-duty passenger vehicles. Furthermore, the union of the two modal systems has not been used for model development. VSP binning remains the most prevalently used on-road data categorization technique to date. However, the nuances found in the VSP/drive-mode combined system are useful discussion points. For example, high VSP cruising results in higher average emissions than low VSP cruising, suggesting that emissions during cruise mode will typically be higher at higher speeds or under conditions of higher engine load (i.e. higher relative grade).
While the VSP binning approach is a more detailed and useful analysis tool, the driving mode methodology is more intuitive and easily understood. For this reason, both VSP binning and driving mode designations were used to analyze the PHEV dataset.