Correlation Coefficients
This section presents the various activities performed and the main results, to estimate the WLTP/NEDC correlation factors.
The activities herein presented target the extraction of WLTP/NEDC correlation factors using the data and the simulation tools that are available within the JRC CO2MPAS team. The main purpose of this exercise is to provide comments on the Determining Equivalent Targets memo of 05 Feb. 2016 prepared by ___, and more specifically on the table entitled "NEDC-WLTP correlation factors for TTW CO2 emissions for all drivetrain technologies and segments assessed for both passenger cars and LCVs".
In order to reach the above target, two separate strategies / methodologies have been identified and pursued separately, in order to estimate in-house correlation factors and then compare them with the ones reported.
The main idea here was to use the data derived from the different simulations / segments used in the Green Driving tool and extract the required correlation factors. The key reasoning was to use the vast amount of data that was already available, including both conventional and electrics / electrified vehicles for the various segments. Although, two issues arose:
- Available data was representative of real word driving and not "standardized tests".
Solution : Use the calibrated Kriging Metamodel and re-run the simulation plans for the different segments with "standardized inputs", e.g. slope and auxiliaries losses (Air Condition) set to zero, taking advantage of the much faster simulation times of the metamodel as compared to CO2MPAS. Under that scope the kriging model was implemented in python as described in the relative section here.
- No data required for the simulation of the electrics / electrified vehicles was available for the NEDC.
The existing simulation plan outputs from CO2MPAS included the required parameters, but the later were not among the selected data which were used for the MATLAB Kriging model calibration.
Solution : Towards that end, the data were reformated to include all the missing elements relative to the NEDC. Resulting data can be found here (View Raw).
- Data available for the simulation of the electrics / electrified vehicles on NEDC not adequate.
After having reformated and used the data of NEDC to calculate CO2 emissions for electric / electrified vehicles it was made clear that the model used for that purpose is highly sensitive on the duration of each mini-phase, with deteriorating results for longer sub-cycle durations. Indeed, since the respective model constitutes a simplification of "Claudio's model for electric vehicles", built specifically for the needs of the Green Driving tool, it does not simulate the state of charge (SOC) evolution throughout the cycle, calculates the delta SOC based on the time percentage where power is positive / negative during the cycle, and thus is really sensitive on the cycle duration.
Solution : Mini-phases were defined for NEDC - as demonstrated bellow - allowing for a division of the complete cycle to shorter parts where the error of the electrics / electrified vehicles model becomes negligible; as demonstrated for WLTP.
Use of the Web Database and the Parametric CO2MPAS Model. [On-going]
The idea was to use the data available from the web database and extract the coefficients for the various segments. Since electrics / electrified vehicles can dramatically affect the results / the coefficient factors for the various penetration scenarios, it is of crucial importance to include electrics / electrified vehicles in the analysis.
The same approach with the one used for the Green Driving Tool was used here. Each individual car of the Web Database has been simulated as a hybrid, plugin-hybrid, and electric, using "Claudio's model for electric vehicles" as described in the relative section of the Green Driving tool. Both cycles, WLTP and NEDC, have been divided into short duration mini-phases allowing for an accurate estimation of their behavior / of the evaluation of the SOC throughout the complete cycle. The two figures below present the division of the two cycles, while the relative data are accessible here. It has to be noted that the mini-phases defined here for WLTP are not the ones used for the Green Driving tool since the scope is different: use of the complete cycle, as opposed to the use of each phase; they have been altered and defined as a sequence with a total duration equal to the one of the legislated cycle to facilitate the calculation of the overall CO2 emission over WLTP of electrics / electrified vehicles.
The target of the exercise is to use the results of the Parametric runs on the WEB database and merge those with the EEA database in order to derive an "assigned registrations number" for each row. The later can be used in order to calculate fleet based correlation coefficients, along with the effect of the WLTP introduction in the Fleet line of CO2 Emissions vs. Mass.
It has to be noted that, as compared with the previous paragraph, at a fleet level, the analysis is performed only for conventional vehicles - not electrics/electrified vehicles - and only for gasoline and diesel fuels since, as described in the GIST link bellow, those represent the vast majority of the European Fleet in 2015.
The description of the main work and the code is available here, while the raw ipython file is located in the Dropbox folder join_dbs. On the section bellow, "Fleet Analysis - Merging with EEA", there is a presentation of the main findings per different vehicle category.
A pool of 26 tested cars is available, on both NEDC and WLTP TMH configurations. This pool represents a starting point for a quick evaluation of the correlation coefficients, though it cannot be considered sufficient in numbers to be representation of the various fleet segments/categories. Additionally, no HEVs & PHEVs are available.
The source excel file of the measured cars is available here (private), while a summary of the resulting ratios is given bellow.
Web Database & Parametric CO2MPAS Model
The Web Database results are used for the extraction of the correlation coefficients.
A summary of the coefficients for the different categories, followed by figures for each individual drive-trains, are provided bellow.
As it can be seen, the coefficients for hybrids are increasing with the size of the vehicle and are much higher than the conventional cars (most probably due to a kind of optimization of the vehicles towards NEDC). This is in agreement with what is presented in the document Determining Equivalent Targets memo of 05 Feb. 2016.
It has to be noted that the PHEVs refer only to the charge sustaining mode, in which the increase in CO2 of the WLTP is more or less compensated by the increased load due to the battery and by the increase need to recharge the battery in the NEDC with respect to WLTP (WHY?). A way to calculate the overall PHEV CO2, including the charge depleting part, still needs to be defined.
Bellow the main results of the exercise for the fleet analysis are given. More detailed results can be found here, while a further discussion of the results is yet to be made.
On the results bellow the clustering method of the Affinity Propagation is used. Samples for the calibration of the model are chosen after optimization which targets zero loss of registrations. Every time that the code is run results are affected due to a different choice of sample thus different clustering. A method to keep the samples constant needs yet to be defined.
CO2 vs Mass Fleet Lines for WLTP & NEDC
The functions for the trendlines need to be re-positioned to be clearly visible.
Summary Results for Avg. and Fleet / Sales Weighted Emissions
