wltp: generate WLTC gear-shifts based on vehicle characteristics

automotive, vehicles, Light-Duty-Vehicles, cars, gears, gearshifs, simulator, driving, engine, WLTP, wltc, UNECE, standards, emissions, fuel-consumption, nedc, python, vehicle
pip install wltp==0.0.9a4


wltp: generate WLTC gear-shifts based on vehicle characteristics

versions: Latest version in PyPI Latest version in Anaconda cloud Latest version in GitHub Version grafted in project's package coordinates Release date grafted in project's package coordinates (build-version: |release|, build-date: |today|) Development Status Supported Python versions of latest release in PyPi Supported conda platforms
documentation: https://wltp.readthedocs.org/
Documentation status
live-demo: JupyterLab for WLTP (stable) JupyterLab for WLTP (dev)
sources: https://github.com/JRCSTU/wltp
Travis continuous integration testing ok? (Linux) Appveyor continuous integration testing ok? (Windows) PyPi downloads Code Style
Github watchers Github stargazers Github forks Issues count
keywords: UNECE, automotive, car, cars, driving, engine, emissions, fuel-consumption, gears, gearshifts, rpm, simulation, simulator, standard, vehicle, vehicles, WLTC, NEDC
Copyright: 2013-2020 European Commission (JRC-IET)
EUPL 1.1+

A python-3.6+ package to generate the gear-shifts of Light-duty vehicles running the :term:`WLTP` driving-cycles, according to :term:`UNECE`'s :term:`GTR`s.


Figure 1: :ref:`annex-2:cycles` for class-3b Vehicles


This wltp python project is still in alpha stage, in the sense that its results are not "correct" by the standard, and no WLTP dyno-tests should rely currently on them.

Some of the known limitations are described in these places:

  • In the :doc:`CHANGES`.
  • Compare results with AccDB in Notebooks/CarsDB-compare.ipynb notebook; launch your private demo-server (JupyterLab for WLTP (stable)) to view it.



The calculator accepts as input the vehicle's technical data, along with parameters for modifying the execution of the :term:`WLTC` cycle, and it then spits-out the gear-shifts of the vehicle, the attained speed-profile, and any warnings. It does not calculate any CO2 emissions.

An "execution" or a "run" of an experiment is depicted in the following diagram:

        .-----------------.                         .------------------.
        :      Input      :                         :      Output      :
        ;-----------------;                         ;------------------;
       ; +--test_mass    ;     ____________        ; +--pmr           ;
      ;  +--n_idle      ;     |            |      ;  +--wltc_class   ;
     ;   +--f0,f1,f2   ;  ==> |   Cycle    | ==> ;   +--...         ;
    ;    +--wot/      ;       | Generator  |    ;    +--cycle      ;
   ;         +--     ;        |____________|   ;     |    +--     ;
  ;      +--n2vs    ;                         ;      +--gwots    ;
 ;           +--   ;                         ;            +--   ;
'-----------------'                         '------------------'

The Input, Output and all its contents are instances of :term:`datamodel` (trees of strings, numbers & pandas objects)


  • Launch the example jupyter notebooks in a private demo server (JupyterLab for WLTP (stable)).
  • Otherwise, install it locally, preferably from the sources (instructions below).
  • pip install :abbr:`"extras" (e.g. pip install wltp[all])`:
    • plot, excel, all, dev, notebook, test, doc


Python-3.6+ is required and Python-3.7 or Python-3.8 recommended. It requires numpy/scipy and pandas libraries with native backends.


On Windows, it is preferable to use the miniconda distribution; although its conda command adds another layer of complexity on top of pip, unlike standard Python, it has pre-built all native libraries required (e.g. numpy/scipy and pandas).

If nevertheless you choose the standard Python, and some packages fail to build when pip-installing them, download these packages from Gohlke's "Unofficial Windows Binaries" and install them manually with:

pip install <package-file-v1.2.3.whl>


Download the sources,


From within the project directory, run one of these commands to install it:

  • for standard python, installing with pip is enough (but might):

    pip install -e .[test]
  • for conda, prefer to install the conda-packages listed in :file:`Notebooks/conda/conda-reqs.txt`, before running the same pip command, like this:

    conda install  --override-channels -c ankostis -c conda-forge -c defaults --file Notebooks/conda/conda-reqs.txt
    pip install -e .[dev]
  • Check installation:

      $ wltp --version
      $ wltp --help
    See: :ref:`wltp-usage`
  • Recreate jupyter notebooks from the paired *.py "py:percent" files (only these files are stored in git-repo), by executing the bash-script:

  • Run pyalgo on all AccDB cars to re-create the H5 file needed for CarsDB-compare notebook, etc:



import pandas as pd
from wltp import datamodel
from wltp.experiment import Experiment

inp_mdl = datamodel.get_model_base()
    "unladen_mass": None,
    "test_mass": 1100,  # in kg
    "p_rated": 95.3,  # in kW
    "n_rated": 3000,  # in RPM
    "n_idle": 600,
    "n2v_ratios": [122.88, 75.12, 50.06, 38.26, 33.63],

    ## For giving absolute P numbers,
    #  rename `p_norm` column to `p`.
    "wot": pd.DataFrame(
        [[600, 0.1],
        [2500, 1],
        [3500, 1],
        [5000, 0.7]], columns=["n", "p_norm"]
    'f0': 395.78,
    'f1': 0,
    'f2': 0.15,
datamodel.validate_model(inp_mdl, additional_properties=True)
exp = Experiment(inp_mdl, skip_model_validation=True)

# exp = Experiment(inp_mdl)
out_mdl = exp.run()
print(f"Available values: \n{list(out_mdl.keys())}")
print(f"Cycle: \n{out_mdl['cycle']}")

See: :ref:`python-usage`

Project files and folders

The files and folders of the project are listed below (see also :ref:`architecture:Architecture`):

+--bin/                     # (shell-scripts) Utilities & preprocessing of WLTC data on GTR and the wltp_db
|   +--bumpver.py           # (script) Update project's version-string
+--wltp/                    # (package) python-code of the calculator
|   +--cycles/              # (package) code & data for the WLTC data
|   +--experiment           # top-level code running the algo
|   +--datamodel            # schemas & defaults for data of algo
|   +--cycler               # code for generating the cycle
|   +--engine               # formulae for engine power & revolutions and gear-box
|   +--vehicle              # formulae for cycle/vehicle dynamics
|   +--vmax                 # formulae estimating `v_max` from wot
|   +--downscale            # formulae downscaling cycles based on pmr/test_mass ratio
|   +--invariants           # definitions & idempotent formulae for physics/engineering
|   +--io                   # utilities for starting-up, parsing, naming and spitting data
|   +--utils                # software utils unrelated to physics or engineering
|   +--cli                  # (OUTDATED) command-line entry-point for launching this wltp tool
|   +--plots                # (OUTDATED) code for plotting diagrams related to wltp cycles & results
|   +--idgears              # (OUTDATED) reconstructs the gears-profile by identifying the actual gears
+--tests/                   # (package) Test-TestCases
    +--vehdb                # Utils for manipulating h5db with accdb & pyalgo cases.
+--docs/                    # (folder) documentation
|   +--pyplots/             # (DEPRECATED by notebooks) scripts plotting the metric diagrams embedded in the README
+--Notebooks/               # Jupyter notebooks for running & comparing results (see `Notebooks/README.md`)
    +--AccDB_src/           # AccDB code & queries extracted and stored as text
+--setup.py                 # (script) The entry point for `setuptools`, installing, testing, etc
+--requirements/            # (txt-files) Various pip-dependencies for tools.


Python usage

First run :command:`python` or :command:`ipython` :abbr:`REPL (Read-Eval-Print Loop)` and try to import the project to check its version:

>>> import wltp

>>> wltp.__version__            ## Check version once more.

>>> wltp.__file__               ## To check where it was installed.         # doctest: +SKIP

If everything works, create the :term:`datamodel` of the experiment. You can assemble the model-tree by the use of:

For instance:

>>> from wltp import datamodel
>>> from wltp.experiment import Experiment

>>> mdl = {
...     "unladen_mass": 1430,
...     "test_mass":    1500,
...     "v_max":        195,
...     "p_rated":      100,
...     "n_rated":      5450,
...     "n_idle":       950,
...     "n_min":        None,                           ## Manufacturers my override it
...     "n2v_ratios":         [120.5, 75, 50, 43, 37, 32],
...     "f0":   100,
...     "f1":   0.5,
...     "f2":   0.04,
... }
>>> mdl = datamodel.upd_default_load_curve(mdl)                   ## need some WOT

For information on the accepted model-data, check the :ref:`code:Schemas`:

>>> from wltp import utils
>>> utils.yaml_dumps(datamodel.model_schema(), indent=2)                                # doctest: +SKIP
$schema: http://json-schema.org/draft-07/schema#
$id: /wltc
title: WLTC data
type: object
additionalProperties: false
- classes

You then have to feed this model-tree to the :class:`~wltp.experiment.Experiment` constructor. Internally the :class:`pandalone.pandel.Pandel` resolves URIs, fills-in default values and validates the data based on the project's pre-defined :term:`JSON-schema`:

>>> processor = Experiment(mdl)         ## Fills-in defaults and Validates model.

Assuming validation passes without errors, you can now inspect the defaulted-model before running the experiment:

>>> mdl = processor.model               ## Returns the validated model with filled-in defaults.
>>> sorted(mdl)                         ## The "defaulted" model now includes the `params` branch.
['driver_mass', 'f0', 'f1', 'f2', 'f_dsc_decimals', 'f_dsc_threshold', 'f_inertial',
 'f_n_clutch_gear2', 'f_n_min', 'f_n_min_gear2', 'f_running_threshold', 'f_safety_margin',
 'f_up_threshold', 'n2v_ratios', 'n_idle', 'n_min_drive1', 'n_min_drive2', 'n_min_drive2_stopdecel',
 'n_min_drive2_up', 'n_min_drive_down', 'n_min_drive_down_start', 'n_min_drive_set',
 'n_min_drive_up', 'n_min_drive_up_start', 'n_rated', 'p_rated', 't_cold_end', 'test_mass',
 'unladen_mass', 'v_cap', 'v_max', 'v_stopped_threshold', 'wltc_data', 'wot']

Now you can run the experiment:

>>> mdl = processor.run()               ## Runs experiment and augments the model with results.
>>> sorted(mdl)                         ## Print the top-branches of the "augmented" model.
[`cycle`, 'driver_mass', 'f0', 'f1', 'f2', `f_dsc`, 'f_dsc_decimals', `f_dsc_raw`,
 'f_dsc_threshold', 'f_inertial', 'f_n_clutch_gear2', 'f_n_min', 'f_n_min_gear2',
 'f_running_threshold', 'f_safety_margin', 'f_up_threshold', `g_vmax`, `is_n_lim_vmax`,
 'n2v_ratios', `n95_high`, `n95_low`, 'n_idle', `n_max`, `n_max1`, `n_max2`, `n_max3`,
 'n_min_drive1', 'n_min_drive2', 'n_min_drive2_stopdecel', 'n_min_drive2_up', 'n_min_drive_down',
 'n_min_drive_down_start', 'n_min_drive_set', 'n_min_drive_up', 'n_min_drive_up_start',
 'n_rated', `n_vmax`, 'p_rated', `pmr`, 't_cold_end', 'test_mass', 'unladen_mass', 'v_cap',
 'v_max', 'v_stopped_threshold', `wltc_class`, 'wltc_data', 'wot', `wots_vmax`]

To access the time-based cycle-results it is better to use a :class:`pandas.DataFrame`:

>>> import pandas as pd, wltp.cycler as cycler, wltp.io as wio
>>> df = pd.DataFrame(mdl['cycle']); df.index.name = 't'
>>> df.shape                            ## ROWS(time-steps) X COLUMNS.
(1801, 107)
>>> wio.flatten_columns(df.columns)
['t', 'V_cycle', 'v_target', 'a', 'phase_1', 'phase_2', 'phase_3', 'phase_4', 'accel_raw',
 'run', 'stop', 'accel', 'cruise', 'decel', 'initaccel', 'stopdecel', 'up', 'p_inert', 'n/g1',
 'n/g2', 'n/g3', 'n/g4', 'n/g5', 'n/g6', 'n_norm/g1', 'n_norm/g2', 'n_norm/g3', 'n_norm/g4',
 'n_norm/g5', 'n_norm/g6', 'p/g1', 'p/g2', 'p/g3', 'p/g4', 'p/g5', 'p/g6', 'p_avail/g1',
 'p_avail/g2', 'p_avail/g3', 'p_avail/g4', 'p_avail/g5', 'p_avail/g6', 'p_avail_stable/g1',
 'p_avail_stable/g2', 'p_avail_stable/g3', 'p_avail_stable/g4', 'p_avail_stable/g5',
 'p_avail_stable/g6', 'p_norm/g1', 'p_norm/g2', 'p_norm/g3', 'p_norm/g4', 'p_norm/g5',
 'p_norm/g6', 'p_resist', 'p_req', 'P_remain/g1', 'P_remain/g2', 'P_remain/g3',
 'P_remain/g4', 'P_remain/g5', 'P_remain/g6', 'ok_p/g3', 'ok_p/g4', 'ok_p/g5', 'ok_p/g6',
 'ok_gear0/g0', 'ok_max_n/g1', 'ok_max_n/g2', 'ok_max_n/g3', 'ok_max_n/g4', 'ok_max_n/g5',
 'ok_max_n/g6', 'ok_min_n_g1/g1', 'ok_min_n_g1_initaccel/g1', 'ok_min_n_g2/g2',
 'ok_min_n_g2_stopdecel/g2', 'ok_min_n_g3plus_dns/g3', 'ok_min_n_g3plus_dns/g4',
 'ok_min_n_g3plus_dns/g5', 'ok_min_n_g3plus_dns/g6', 'ok_min_n_g3plus_ups/g3',
 'ok_min_n_g3plus_ups/g4', 'ok_min_n_g3plus_ups/g5', 'ok_min_n_g3plus_ups/g6', 'ok_n/g1',
 'ok_n/g2', 'ok_n/g3', 'ok_n/g4', 'ok_n/g5', 'ok_n/g6', 'ok_gear/g0', 'ok_gear/g1',
 'ok_gear/g2', 'ok_gear/g3', 'ok_gear/g4', 'ok_gear/g5', 'ok_gear/g6', 'G_scala/g0', 'G_scala/g1',
 'G_scala/g2', 'G_scala/g3', 'G_scala/g4', 'G_scala/g5', 'G_scala/g6', 'g_min', 'g_max0']
>>> 'Mean engine_speed: %s' % df.n.mean()                                       # doctest: +SKIP
'Mean engine_speed: 1908.9266796224322'
>>> df.describe()                                                               # doctest: +SKIP
           v_class     v_target  ...     rpm_norm       v_real
count  1801.000000  1801.000000  ...  1801.000000  1801.000000
mean     46.361410    46.361410  ...     0.209621    50.235126
std      36.107745    36.107745  ...     0.192395    32.317776
min       0.000000     0.000000  ...    -0.205756     0.200000
25%      17.700000    17.700000  ...     0.083889    28.100000
50%      41.300000    41.300000  ...     0.167778    41.300000
75%      69.100000    69.100000  ...     0.285556    69.100000
max     131.300000   131.300000  ...     0.722578   131.300000
[8 rows x 10 columns]

>>> processor.driveability_report()                                             # doctest: +SKIP
  12: (a: X-->0)
  13: g1: Revolutions too low!
  14: g1: Revolutions too low!
  30: (b2(2): 5-->4)
  38: (c1: 4-->3)
  39: (c1: 4-->3)
  40: Rule e or g missed downshift(40: 4-->3) in acceleration?
  42: Rule e or g missed downshift(42: 3-->2) in acceleration?

You can export the cycle-run results in a CSV-file with the following pandas command:

>>> df.to_csv('cycle.csv')                                                      # doctest: +SKIP

For more examples, download the sources and check the test-cases found under the :file:`/tests/` folder.

Cmd-line usage


Not implemented in yet.

The command-line usage below requires the Python environment to be installed, and provides for executing an experiment directly from the OS's shell (i.e. :program:`cmd` in windows or :program:`bash` in POSIX), and in a single command. To have precise control over the inputs and outputs (i.e. experiments in a "batch" and/or in a design of experiments) you have to run the experiments using the API python, as explained below.

The entry-point script is called :program:`wltp`, and it must have been placed in your :envvar:`PATH` during installation. This script can construct a model by reading input-data from multiple files and/or overriding specific single-value items. Conversely, it can output multiple parts of the resulting-model into files.

To get help for this script, use the following commands:

$ wltp --help                               ## to get generic help for cmd-line syntax
$ wltcmdp.py -M vehicle/full_load_curve     ## to get help for specific model-paths

and then, assuming vehicle.csv is a CSV file with the vehicle parameters for which you want to override the n_idle only, run the following:

$ wltp -v \
    -I vehicle.csv file_frmt=SERIES model_path=params header@=None \
    -m vehicle/n_idle:=850 \
    -O cycle.csv model_path=cycle

Excel usage


OUTDATED!!! Excel-integration requires Python 3 and Windows or OS X!

In Windows and OS X you may utilize the excellent xlwings library to use Excel files for providing input and output to the experiment.

To create the necessary template-files in your current-directory you should enter:

$ wltp --excel

You could type instead :samp:`wltp --excel {file_path}` to specify a different destination path.

In windows/OS X you can type :samp:`wltp --excelrun` and the files will be created in your home-directory and the excel will open them in one-shot.

All the above commands creates two files:


The python-enabled excel-file where input and output data are written, as seen in the screenshot below:

Screenshot of the `wltp_excel_runner.xlsm` file.

After opening it the first tie, enable the macros on the workbook, select the python-code at the left and click the :menuselection:`Run Selection as Python` button; one sheet per vehicle should be created.

The excel-file contains additionally appropriate VBA modules allowing you to invoke Python code present in selected cells with a click of a button, and python-functions declared in the python-script, below, using the mypy namespace.

To add more input-columns, you need to set as column Headers the json-pointers path of the desired model item (see :ref:`python-usage` below,).


Utility python functions used by the above xls-file for running a batch of experiments.

The particular functions included reads multiple vehicles from the input table with various vehicle characteristics and/or experiment parameters, and then it adds a new worksheet containing the cycle-run of each vehicle . Of course you can edit it to further fit your needs.


You may reverse the procedure described above and run the python-script instead. The script will open the excel-file, run the experiments and add the new sheets, but in case any errors occur, this time you can debug them, if you had executed the script through LiClipse, or IPython!

Some general notes regarding the python-code from excel-cells:

  • On each invocation, the predefined VBA module pandalon executes a dynamically generated python-script file in the same folder where the excel-file resides, which, among others, imports the "sister" python-script file. You can read & modify the sister python-script to import libraries such as 'numpy' and 'pandas', or pre-define utility python functions.
  • The name of the sister python-script is automatically calculated from the name of the Excel-file, and it must be valid as a python module-name. Therefore do not use non-alphanumeric characters such as spaces(`` ), dashes(-) and dots(.``) on the Excel-file.
  • On errors, a log-file is written in the same folder where the excel-file resides, for as long as the message-box is visible, and it is deleted automatically after you click 'ok'!
  • Read http://docs.xlwings.org/quickstart.html


The Python code is highly modular, with testability in mind. so that specific parts can run in isolation. This facilitates studying tough issues, such as, double-precision reproducibility, boundary conditions, comparison of numeric outputs, and studying the code in sub-routines.


Run test-cases with pytest command.

Data Structures:

Computations are vectorial, based on hierarchical dataframes, all of them stored in a single structure, the datamodel. In case the computation breaks, you can still retrieve all intermediate results till that point.

.. TODO::
    Almost all of the names of the `datamodel` and `formulae` can be remapped,
    For instance, it is possible to run the tool on data containing ``n_idling_speed``
    instead of ``n_idle`` (which is the default), without renaming the input data.

.. glossary::

        The container of all the scalar Input & Output values, the WLTC constants factors,
        and 3 matrices: `WOT`, `gwots`, and the `cycle run` time series.

        It is composed by a stack of mergeable `JSON-schema` abiding trees of *string, numbers & pandas objects*,
        formed with python *sequences & dictionaries, and URI-references*.
        It is implemented in :mod:`~wltp.datamodel`, supported by :class:`pandalone.pandata.Pandel`.

    Full Load Curve
        An *input* array/dict/dataframe with the full load power curves for (at least) 2 columns for ``(n, p)``
        or their normalized values ``(n_norm, p_norm)``.
        See also https://en.wikipedia.org/wiki/Wide_open_throttle

    grid WOTs
        A dataframe produced from `WOT` for all gear-ratios, indexed by a grid of rounded velocities,
        and with 2-level columns ``(item, gear)``.
        It is generated by :func:`~wltp.engine.interpolate_wot_on_v_grid()`, and augmented
        by :func:`~wltp.engine.attach_p_avail_in_gwots()` & :func:`~wltp.vehicle.calc_p_resist()` .

        .. TODO::
            Move `grid WOTs` code in own module :mod:`~wltp.gwots`.

    cycle run
        A dataframe with all the time-series, indexed by the time of the samples.
        The velocities for each time-sample must exist in the `gwots`.
        The columns are the same 2-level columns like *gwots*.
        it is implemented in :mod:`~wltp.cycler`.

Code Structure:

The computation code is roughly divided in these python modules:

.. glossary::

        Physics and engineering code, implemented in modules:

        - :mod:`~wltp.engine`
        - :mod:`~wltp.vmax`
        - :mod:`~wltp.downscale`
        - :mod:`~wltp.vehicle`

    - orchestration
        The code producing the actual gear-shifting, implemented in modules:

        - :mod:`~wltp.datamodel`
        - :mod:`~wltp.cycler`
        - :mod:`~wltp.gridwots` (TODO)
        - :mod:`~wltp.scheduler` (TODO)
        - :mod:`~wltp.experiment` (TO BE DROPPED, :mod:`~wltp.datamodel` will assume all functionality)

        The internal software component :mod:`graphtik` which decides which
        `formulae` to execute based on given inputs and requested outputs.

The blueprint for the underlying software ideas is given with this diagram:

Software architectural concepts underlying WLTP code structure.

Note that currently there is no scheduler component, which will allow to execute the tool with a varying list of available inputs & required data, and automatically compute only what is not already given.

Specs & Algorithm

This program imitates to some degree the MS Access DB (as of July 2019), following this 08.07.2019_HS rev2_23072019 GTR specification (:download:`docs/_static/WLTP-GS-TF-41 GTR 15 annex 1 and annex 2 08.07.2019_HS rev2_23072019.docx`, included in the :file:`docs/_static` folder).


There is a distinctive difference between this implementation and the AccDB:

All computations are vectorial, meaning that all intermediate results are calculated & stored, for all time sample-points, and not just the side of the conditions that evaluate to true on each sample.

The latest official version of this GTR, along with other related documents maybe found at UNECE's site:


The WLTC-profiles for the various classes were generated from the tables of the specs above using the :file:`devtools/csvcolumns8to2.py` script, but it still requires an intermediate manual step involving a spreadsheet to copy the table into ands save them as CSV.





Phases (the problem)

GTR's "V" phasings

The :term:`GTR`'s velocity traces have overlapping split-time values, i.e. belonging to 2 phases, and e.g. for class1 these are the sample-values @ times 589 & 1022:

GTR's "V" phasing scheme for Velocities
class1 phase-1 phase-2 phase-3 cycle
Boundaries [0, 589] [589, 1022] [1022, 1611] [0, 1611]
Duration 589 433 589 1611
# of samples 590 434 590 1612

"Semi-VA" phasings

Some programs and most spreadsheets do not handle overlapping split-time values like that (i.e. keeping a separate column for each class-phase), and assign split-times either to the earlier or the later phase, distorting thus the duration & number of time samples some phases contain!

For instance, Access-DB tables assign split-times on the lower parts, distorting the start-times & durations for all phases except the 1st one (deviations from GTR in bold):

Access-DB, a "semi-VA1" phasing scheme (all but 1st phases shorter)
class1 phase-1 phase-2 phase-3 cycle
Boundaries [0, 589] [590, 1022] [1023, 1611] [0, 1611]
Duration 589 432 588 1611
# of samples 590 433 589 1612


The algorithms contained in Access DB are carefully crafted to do the right thing.

The inverse distortion (assigning split-times on the higher parts) would preserve phase starting times (hint: downscaling algorithm depends on those absolute timings being precisely correct):

"Inverted" Access-DB, a "semi-VA0" phasing schema (all but last phases shorter)
class1 phase-1 phase-2 phase-3 cycle
Boundaries [0, 588] [589, 1021] [1022, 1611] [0, 1611]
Duration 588 432 589 1611
# of samples 589 433 590 1612

"VA" phasings

On a related issue, GTR's formula for Acceleration (Annex 1 3.1) produces one less value than the number of velocity samples (like the majority of the distorted phases above). GTR prescribes to (optionally) append and extra A=0 sample at the end, to equalize Acceleration & Velocities lengths, but that is not totally ok (hint: mean Acceleration values do not add up like mean-Velocities do, see next point about averaging).

Since most calculated and measured quantities (like cycle Power) are tied to the acceleration, we could refrain from adding the extra 0, and leave all phases with -1 samples, without any overlapping split-times:

"VA0" phasings
class1 phase-1 phase-2 phase-3 cycle
Boundaries [0, 588] [589, 1021] [1022, 1610] [0, 1610]
Duration 588 432 588 1610
# of samples 589 433 589 1611

Actually this is "semi-VA0" phasings, above, with the last part equally distorted by -1 @ 1610. But now the whole cycle has (disturbingly) -1 # of samples & duration:

We can resolve this, conceptually, by assuming that each Acceleration-dependent sample signifies a time-duration, so that although the # of samples are still -1, the phase & cycle durations (in sec) are as expected:

"VA0+" phasings, with 1 sec step duration
class1 phase-1 phase-2 phase-3 cycle
Boundaries [0, 589 ) [589, 1022 ) [1022, 1611 ) [0, 1611 )
Duration 589 433 589 1611
# of samples 589 433 589 1611

Summarizing the last "VA0+" phasing scheme:

  • each step signifies a "duration" of 1 sec,
  • the duration of the final sample @ 1610 reaches just before 1611sec,
  • # of samples for all phases are symmetrically -1 compared to Velocity phases,
  • it is valid for Acceleration-dependent quantities only,
  • it is valid for any sampling frequency (not just 1Hz),
  • respects the Dijkstra counting (notice the parenthesis signifying open right intervals, in the above table), BUT ...
  • Velocity-related quantities cannot utilize this phasing scheme, must stick to the original, with overlapping split-times.

Averaging over phases

Calculating mean values for Acceleration-related quantities produce correct results only with non-overlapping split-times.

It's easier to demonstrate the issues with a hypothetical 4-sec cycle, composed of 2 symmetrical ramp-up/ramp-down 2-sec phases (the "blue" line in the plot, below):

ramp-up/down cycle
t V-phase1 V-phase2 V Distance VA-phase A
[sec]     [kmh] [m x 3.6]   [m/sec²]
0 X   0 0 1 5
1 X   5 2.5 1 5
2 X X 10 10 2 -5
3   X 5 17.5 2 -5
4   X 0 20 <blank> <blank>
  • The final A value has been kept blank, so that mean values per-phase add up, and phases no longer overlap.
mean values for ramp-up/down cycle, above
mean(V) mean(S) mean(A)
[kmh] [m x 3.6] [m/sec²]
phase1: 5 10 5
phase2: 5 10 -5
  • Applying the V-phasings and the extra 0 on mean(A) would have arrived to counterintuitive values, that don't even sum up to 0:
    • up-ramp: \left(\frac{5 + 5 + (-5)}{3} =\right) 1.66m/sec^2
    • down-ramp: \left(\frac{(-5) + (-5) + 0}{3} =\right) -3.33m/sec^2

Practical deliberations

All phases in WLTC begin and finish with consecutive zeros(0), therefore the deliberations above do not manifest as problems; but at the same time, discovering off-by-one errors & shifts in time (wherever this really matters e.g. for syncing data), on arbitrary files containing calculated and/or measured traces is really hard: SUMs & CUMSUMs do not produce any difference at all.

The tables in the next section, along with accompanying CRC functions developed in Python, come as an aid to the problems above.

Phase boundaries

As reported by :func:`wltp.cycles.cycle_phases()`, and neglecting the advice to optionally add a final 0 when calculating the cycle Acceleration (Annex 1 2-3.1), the following 3 phasing are identified from velocity traces of 1Hz:

  • V: phases for quantities dependent on Velocity samples, overlapping split-times.
  • VA0: phases for Acceleration-dependent quantities, -1 length, NON overlapping split-times, starting on t=0.
  • VA1: phases for Acceleration-dependent quantities, -1 length, NON overlapping split-times, starting on t=1. (e.g. Energy in Annex 7).
class phasing phase-1 phase-2 phase-3 phase-4
class1 V [0, 589] [589, 1022] [1022, 1611]  
VA0 [0, 588] [589, 1021] [1022, 1610]  
VA1 [1, 589] [590, 1022] [1023, 1611]  
class2 V [0, 589] [589, 1022] [1022, 1477] [1477, 1800]
VA0 [0, 588] [589, 1021] [1022, 1476] [1477, 1799]
VA1 [1, 589] [590, 1022] [1023, 1477] [1478, 1800]
class3a V [0, 589] [589, 1022] [1022, 1477] [1477, 1800]
VA0 [0, 588] [589, 1021] [1022, 1476] [1477, 1799]
VA1 [1, 589] [590, 1022] [1023, 1477] [1478, 1800]
class3b V [0, 589] [589, 1022] [1022, 1477] [1477, 1800]
VA0 [0, 588] [589, 1021] [1022, 1476] [1477, 1799]
VA1 [1, 589] [590, 1022] [1023, 1477] [1478, 1800]


CRCs & CUMSUMs for all phases over different "phasings"
by_phase cumulative by_phase cumulative
phasing⇨ phase⇩ V VA0 VA1 V VA0 VA1 V V
phase1 9840 4438 97DB 9840 4438 97DB 11988.4 11988.4
phase2 8C34 8C8D D9E8 DCF2 090B 4295 17162.8 29151.2
phase3 9840 4438 97DB 6D1D 4691 F523 11988.4 41139.6
phase1 8591 CDD1 8A0A 8591 CDD1 8A0A 11162.2 11162.2
phase2 312D 391A 64F1 A010 606E 3E77 17054.3 28216.5
phase3 81CD E29E 9560 28FB 9261 D162 24450.6 52667.1
phase4 8994 0D25 2181 474B 262A F70F 28869.8 81536.9
phase1 48E5 910C 477E 48E5 910C 477E 11140.3 11140.3
phase2 1494 D93B 4148 403D 2487 DE5A 16995.7 28136.0
phase3 8B3B 9887 9F96 D770 3F67 2EE9 25646.0 53782.0
phase4 F962 1A0A 5177 9BCE 9853 2B8A 29714.9 83496.9
phase1 48E5 910C 477E 48E5 910C 477E 11140.3 11140.3
phase2 AF1D E501 FAC1 FBB4 18BD 65D3 17121.2 28261.5
phase3 15F6 A779 015B 43BC B997 BA25 25782.2 54043.7
phase4 F962 1A0A 5177 639B 0B7A D3DF 29714.9 83758.6

... where if a some cycle-phase is identified as:

  • V phasing, it contains all samples;
  • VA0 phasing, it lacks -1 sample from the end;
  • VA1 phasing, it lacks -1 sample from the start.

Practical example

For instance, let's identify the V-trace of class1's full cycle:

>>> from wltp.datamodel import get_class_v_cycle  as wltc
>>> from wltp.cycles import identify_cycle_v as crc
>>> V = wltc("class1")
>>> crc(V)            # full cycle
('class1', None, 'V')
>>> crc(V[:-1])       # -1 (last) sample
('class1', None, 'VA0')

The :func:`crc() <wltp.cycles.crc_velocity>` function returns a 3-tuple: (i-class, i-phase, i-kind):

  • When i-phase is None, the trace was a full-cycle.

Now let's identify the phases of "Access-DB":

>>> crc(V[:590])      # AccDB phase1 respects GTR
('class1', 'phase-1', 'V')
>>> crc(V[590:1023])  # AccDB phase2 has -1 (first) sample
('class1', 'phase-2', 'VA1')
>>> crc(V[:1023])     # cumulative AccDB phase2 respects GTR
('class1', 'PHASE-2', 'V')
  • When i-phase is CAPITALIZED, the trace was cumulative.
  • Phase2 is missing -1 sample from the start (i-kind == VA1).
>>> crc(V[1023:])     # AccDB phase3
('class1', 'phase-1', 'VA1')
  • Phase3 was identified again as phase1, since they are identical.

Finally, clipping both samples from start & end, matches no CRC:

>>> crc(V[1:-1])
(None, None, None)

To be note, all cases above would have had identical CUMSUM (GTR's) CRCs.

Getting Involved

This project is hosted in github. To provide feedback about bugs and errors or questions and requests for enhancements, use github's Issue-tracker.

Development procedure

For submitting code, use UTF-8 everywhere, unix-eol(LF) and set git --config core.autocrlf = input.

The typical development procedure is like this:

  1. Install and arm a pre-commit hook with black to auto-format you python-code.

  2. Modify the sources in small, isolated and well-defined changes, i.e. adding a single feature, or fixing a specific bug.

  3. Add test-cases "proving" your code.

  4. Rerun all test-cases to ensure that you didn't break anything, and check their coverage remain above the limit set in :file:`setup.cfg`.

  5. If you made a rather important modification, update also the :doc:`CHANGES` file and/or other documents (i.e. README.rst). To see the rendered results of the documents, issue the following commands and read the result html at :file:`build/sphinx/html/index.html`:

    python setup.py build_sphinx                  # Builds html docs
    python setup.py build_sphinx -b doctest       # Checks if python-code embedded in comments runs ok.
  6. If there are no problems, commit your changes with a descriptive message.

  7. Repeat this cycle for other bugs/enhancements.

  8. When you are finished, push the changes upstream to github and make a merge_request. You can check whether your merge-request indeed passed the tests by checking its build-status Travis continuous integration testing ok? (Linux) on the integration-server's site (TravisCI).


    Skim through the small IPython developer's documentation on the matter: The perfect pull request

  9. Generate the Sphinx documents in :file:`./wltp.git/docs/_build/html/` with this command:

    python setup.py build_sphinx

    Access the generated documents through a a web-server, for :term:`graphtik` graphs to work correctly, with this bash-command (remove the final & on Windows):

    python -m http.server 8080 --directory ./wltp.git/docs/_build/html/ &

Development team

  • Author:
    • Kostis Anagnostopoulos
  • Contributing Authors:
    • Heinz Steven (test-data, validation and review)
    • Georgios Fontaras (simulation, physics & engineering support)
    • Alessandro Marotta (policy support)
    • Jelica Pavlovic (policy support)
    • Eckhard Schlichte (discussions & advice)


See also :ref:`architecture:Architecture`.

.. glossary::

        The `Worldwide harmonised Light duty vehicles Test Procedure <https://www2.unece.org/wiki/pages/viewpage.action?pageId=2523179>`_,
        a `GRPE` informal working group

        The United Nations Economic Commission for Europe, which has assumed the steering role
        on the `WLTP`.

        `UNECE` Working party on Pollution and Energy - Transport Programme

        Any of the *Global Technical Regulation* documents of the `WLTP` .

    GS Task-Force
        The Gear-shift Task-force of the `GRPE`. It is the team of automotive experts drafting
        the gear-shifting strategy for vehicles running the `WLTP` cycles.

        The family of pre-defined *driving-cycles* corresponding to vehicles with different
        :abbr:`PMR (Power to Mass Ratio)`. Classes 1,2, 3a/b are split in 3, 4 and 4 *parts* respectively.

    MS Access DB
        The original implementation of the algorithm in *MS Access* by Heinz Steven.

        To facilitate searching and cross-referencing the existing routines,
        all the code & queries of the database have been extracted and stored in as text
        under the `Notebooks/AccDB_src/
        <https://github.com/JRCSTU/wltp/tree/master/Notebooks/AccDB_src/>`_ folder
        of this project.

    Mass in running order
        The mass of the vehicle, with its fuel tank(s) filled to at least 90 per cent
        of its or their capacity/capacities, including the mass of the driver and the liquids,
        fitted with the standard equipment in accordance with the manufacturer’s specifications and,
        where they are fitted, the mass of the bodywork, the cabin,
        the coupling and the spare wheel(s) as well as the tools when they are fitted.

    Kerb mass
    Curb weight
    Unladen mass
        The `Mass in running order` minus the `Driver mass`.

    Driver weight
    Driver mass
        75 kgr

    Test mass
        The representative weight of the vehicle used as input for the calculations of the simulation,
        derived by interpolating between high and low values for the |CO2|-family of the vehicle.

        Reduction of the top-velocity of the original drive trace to be followed, to ensure that the vehicle
        is not driven in an unduly high proportion of "full throttle".

        The `JSON schema <http://json-schema.org/>`_ is an `IETF draft <http://tools.ietf.org/html/draft-zyp-json-schema-03>`_
        that provides a *contract* for what JSON-data is required for a given application and how to interact
        with it.  JSON Schema is intended to define validation, documentation, hyperlink navigation, and
        interaction control of JSON data.

        The schema of this project has its own section: :ref:`code:Schemas`

        You can learn more about it from this `excellent guide <http://spacetelescope.github.io/understanding-json-schema/>`_,
        and experiment with this `on-line validator <http://www.jsonschema.net/>`_.

        JSON Pointer(:rfc:`6901`) defines a string syntax for identifying a specific value within
        a JavaScript Object Notation (JSON) document. It aims to serve the same purpose as *XPath* from the XML world,
        but it is much simpler.

        The text-oriented language, a superset of `Restructured Text <https://en.wikipedia.org/wiki/ReStructuredText>`_,
        used to write the documentation for this project, with similar capabilities to *LaTeX*,
        but for humans, e.g.,  the Linux kernel adopted this textual format on 2016.

    jupyter notebook
        *Jupyter* is a web-based interactive computational environment for creating *Jupyter notebook* documents.
        The "notebook" term can colloquially make reference to many different entities,
        mainly the Jupyter web application, Jupyter Python web server, or Jupyter document format,
        depending on context.

        A *Jupyter Notebook* document is composed of an ordered list of input/output *cells*
        which contain code in various languages, text (using Markdown), mathematics, plots and
        rich media, usually ending with the ".ipynb" extension.