Green Driving
Short description of the tool / To be filled.
To be filled.
Description of the overall methodology: preparation of the database, meta-model, electrics, server.
Description of the work performed to prepare the database, using calibrated models representing the different segments. Segmentation, simulation plan, etc. To be filled.
The available data on the web database were used, along with the Parametric CO2MPAS, to prepare an alternative input database for the calibration of the Kriging meta-model. This targets at addressing the "perceived" limitations of using a unique pre-calibrated model of the CO2MPAS tool as representative of a fleet segment, by using directly the inputs of real cars from the various segments. Additionally, this way the number of the calibrated Kriging meta-models could be limited to a restricted number of pre-defined segments - e.g. manuals or automatics, engine type and fuel - and thus improve both the hurdles of dealing with multiple databases for each segment and different engine types, and the calculation times of the tool.
In order to achieve that, the following two tasks had to be implemented on top of what had already been done for the Web Database Runs.
-
Add two variables representing the slope and the auxiliaries power losses (representing Air Condition On/Off) to the inputs, and assign random numbers to them within the respective limits; Commit: c3226be.
-
Add the complete set of outputs required for the calculation of the electrics / electrified vehicles, for each mini-phase; Commit: a2a9614.
The results of the above work / the Parametric CO2MPAS runs for conventional vehicles, can be found here (results contained here concern the 1,292 cars of the final web database for 2015, and for the various WLTP mini-phases; the last rows represent the whole WLTP cycle for cross checking (as it can be seen from the initial T)), while all results of the calibration of the Kriging meta-model on those are available here: J:\Sustainable Transport Unit\CO2_from_Road_Transport\ComplemetaryActivities\GreenDriving\simulations\mm_6.
For reference, the following picture present the CO2 emission results of the Kriging meta-model for gasoline turbo manual vehicles.
Update
After having identified a limitation of the previous approach for cases outside the strict "calibration space" of the Kriging - more specifically on the Road Loads and on the combinations of mass/F0 and the noise in F2 introduced by the roofbox - , we decide to extend the space by sampling more cars from the web database and adding a slight noise in parameters that should not "affect" the vehicles.
More specifically, the following steps have been performed (Commit: 16840ff:
- Extend the sample with cars from 2014 on-wards; approx. 3,700 cars in total
- Further extend the sample with cars having the min and max capacity, power and mass for each segment (dt_mt, dt_at, gna_mt, etc) from all the database (all years); we choose the 100 cars for each case and we add them to the previously selected sample; this leads to an overall sample of 5,700 cars in total certainly including extreme cases
- Add a noise to the reference mass; sampled from a normal distribution with mean = 1 and stdev = 0.1 (this is multiplied with the reference value)
- Add a random noise to the wheel rolling resistance; can go from 0.0065 to 0.013
- Add a noise to the frontal area; sampled from a normal distribution with mean = 1 and stdev = 0.1 (this is multiplied with the reference value)
- Add a random noise to f1; can go from 1reference value to 2reference value.
- Define min and max limits for each segment and each parameter and ask Lorenzo to implement those on the users' selection; the user should not be able to select input values outside this predefined range; bellow the ranges of the selected db (those may change after the co2mpas runs, e.g. completely wrong results)
Ranges:
| Capacity [cc] | Power [kW] | Mass [kg] | |
|---|---|---|---|
| Segment | min / max | min / max | min / max |
| DT_MT | 799 / 4164 | 30 / 180 | 742.5 / 3105 |
| DT_AT | 799 / 5934 | 30 / 368 | 718 / 3553.5 |
| GNA_MT | 617 / 8277 | 22 / 493 | 498 / 2945.5 |
| GNA_AT | 796 / 6750 | 29 / 515 | 674 / 3455.5 |
| GT_MT | 599 / 3980 | 33 / 340 | 668.5 / 2379.5 |
| GT_AT | 599 / 7993 | 37 / 882 | 730.5 / 3360 |
A first attempt to calibrate meta-models both for conventional (co2_emission) and electric / electrified vehicles (co2_hev & co2_phev) was made in order to avoid running "Claudio's model for electric vehicles" on the server side and use directly a meta-model for both types of vehicles. Though, the Kriging meta-model fails to capture well the results for the electric / electrified vehicles, leading to an alternative approach, as explained below.
The Kriging approach is really sensitive to "non-continuous" outputs, where "steps" are present (Put reference here). As it can be seen from the raw data that is the case for the CO2 emissions from electric / electrified vehicles. This is expected given that the operation of electrics / electrified vehicles is highly affected by the various strategies included, which changes the way some of the outputs are calculated. In order to bypass this problem, all of possible outputs are calculated - as continuous variables - and fed into Kriging for calibration. The various strategies should be implemented at a later stage at the server side.
Results of "Claudio's model for electric vehicles" - the input of Kriging - are provided here.
A script has been prepared to combine inputs and outputs per segment and produce the files to be used for the Kriging meta-model calibration (commit: 515a415). The script is available here, while the results of the script's run / inputs for Kriging meta-model's calibration are available here (private repo).
Move here / describe results.