Mars Express Power Challenge

ECML PKDD 2017
DISCOVERY CHALLENGE


Take part!

It has now been more than 12 years that the Mars Express Orbiter (MEX) Mars Express Orbiter (MEX) provides science data from Mars about its ionosphere and ground subsurface composition. The 3D imagery of Mars has provided the community with unprecedented information about the planet. Today, thanks to the work of careful and expert operators, Mars Express Orbiter still provides information that supports ground exploration missions on Mars (Curiosity, Opportunity, ...) and a lot of other research.

The Mars Express Orbiter is operated by the European Space Agency from its operations center (Darmstadt, Germany) where all the telemetry is analyzed. The health status of the spacecraft is carefully monitored to plan future science observations and to avoid power shortages.

Operators of Mars Express keep track of the thermal power consumption thanks to the telemetry data. The spacecraft uses electric power coming from the solar arrays (or batteries, during eclipses) not only to supply power to the platform units, but also to the thermal subsystem, which keeps the entire spacecraft within its operating temperature range. The remaining available power can be used by the payloads to do science operations:

$$ SciencePower = ProducedPower - Platform Power - ThermalPower$$

The Mars Express Power Challenge focuses on the difficult problem of predicting the thermal power consumption. More than six Earth years of Mars Express telemetry are made available and you are challenged to predict the thermal subsystem power consumption on the following 26 Earth months. If successful, the winning method will be integrated in the new tool helping operators of the Mars Express Orbiter to deliver science data for a longer period of time.

This competition is organized by the ESA Data Analytics Team for Operations and the Mars Express Flight Control Team. Both teams will be available for interactions via the competition discussion channel on slack.

The competition will be also a part of the ECML PKDD 2017 (European Conference on Machine Learning & Principles and Practice of Knowledge Discovery), will take place in Skopje, Macedonia, September 18–22, 2017

Follow the competition with the hashtag #MarsExpressPower

Challenge

The Mars Express spacecraft orbiting around Mars is operated by the European Space Agency (ESA). A dedicated mission control team, working from the ESA Operations Centre (ESOC), prepares the mission plans and sends telecommands to the spacecraft such as:

  • orbit and attitude control
  • science payload ON/OFF commands
  • radio reception and transmission plans
  • other sets of commands, etc.

The spacecraft has an internal thermal subsystem. It works autonomously to keep each part of the spacecraft within pre-defined working temperatures. This thermal subsystem is composed of nodes (including heaters and coolers) that consume electric power as needed. Predicting the behavior of this thermal subsystem by modelling the electric power consumption in each of these nodes is very important to mission operators. This prediction gives direct insight on the heaters' power consumption and is used to provide safer flights with better plans for optimized science operations.

The goal of this challenge is to analyze the provided training set of Mars Express data, including context data (explanatory/predictor variables) and measurements of the electric current (target variables) in each thermal subsystem node, to predict the average electric current of 33 thermal power lines per hour of the test set containing only the context data.

Data

The data for this competition can be downloaded as a compressed zip file.


Download


The ESA Data Analytics Team for Operations has prepared the data with care. If you have any questions after reading this description please feel free to post on the slack channel.

Licence - ESA - CC BY-SA 3.0 IGO

This work is licensed under the Creative Commons Attribution-ShareAlike 3.0 IGO License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/3.0/igo/ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.


Description

The training set (train_set/) is composed of context and electric current measurements data for the following four Martian years:

  • Set A: 2008-08-22 to 2010-07-10
  • Set B: 2010-07-10 to 2012-05-27
  • Set C: 2012-05-27 to 2014-04-14
  • Set D: 2014-04-14 to 2015-01-01

The test set (test_set/) is composed of context data only for the following fifth Martian year:

  • Set E: 2015-01-01 to 2017-03-01

A Martian year is defined by 687 earth days which is not respected by smaller sets D and E. Note that space communication issues may have introduced gaps in the given data.


Context data

Context files are all in CSV format and are using the following nomenclature:

  • context--YYYY-MM-DD__YYYY-MM-DD--TYPE.csv
    • YYYY-MM-DD__YYYY-MM-DD indicates the start and end dates in Earth UTC time.
    • TYPE: there are 5 types of context files:
      • SAAF: Solar Aspect Angles
      • DMOP: Detailed Mission Operations Plan
      • FTL: Flight dynamics TimeLine with spacecraft pointing events
      • EVTF: Other events
      • LTDATA: long term data such as the sun-mars distance

The following list describes the columns of each of these CSV file types:

  • SAAF files: Solar aspect angles are expressed with respect to the Sun-MarsExpress line.

    • ut_ms: unix timestamp in milliseconds
    • sa: solar aspect angle, angle of Mars Express solar panels' normal
    • sx: solar angle of the X axis of satellite
    • sy: solar angle of the Y axis of satellite
    • sz: solar angle of the Z axis of satellite

    The angles are a good information about how much power from the solar panels the spacecraft is going to get (sa) and how much each of its faces is roasting to the sun or freezing in its own shadow (sx, sy, sz).

  • DMOP files: Detailed Mission Operations Plan files.

    • ut_ms: unix timestamp in milliseconds
    • subsystem: name of the operated subsystem command

    The DMOP files show when the different subsystems' commands have been triggered. Each subsystem command string include the subsystem name along with the command name. Each subsystem command name can be considered as a signature. Due to the amount of different commands, they won't be explained here. These commands have different effects on temperatures of the different subsystems of the spacecraft thus impacting in different ways the thermal subsystem's reactions. In these commands you may find the ON/OFF of radio communications, the ON/OFF of the science instruments, etc.

  • FTL files: listing of spacecraft pointing events (Flight Dynamics TimeLine).

    • ute_ms: unix timestamp in milliseconds of the time the pointing period begins
    • ute_ms: unix timestamp in milliseconds of the time the same pointing period ends
    • type: type of pointing or action
    • flagcomms: TRUE if any communication device was used, else FALSE.

    The events listed here are pointing and action commands that can impact the attitude of the satellite, thus they also may impact the solar aspect angles of the orbiter and/or the switch ON/OFF of some instrumentation.

  • EVTF files: more events are listed in these event files.

    • ut_ms: unix timestamp in milliseconds
    • description: short description of the event

    This file contains a lot of events. It covers parts of the flight dynamics timeline present in the FTL and complements information with a lot of other events. In this file you can find the following eclipse events; *_PENUMBRA_START, *_PENUMBRA_END, *_UMBRA_START, *_UMBRA_END which indicates the start and end time of different phases of the eclipse (penumbra/umbra)

    Some of the descriptions are self-explanatory while others are not. Due to the amount of different events, they won't be explained here. Nevertheless, the descriptions can be used as signatures of events, similar to the subsystem commands in the DMOP files.

  • LTDATA files: long term data including sun-mars distance and solar constant on Mars.

    • ut_ms: unix timestamp in milliseconds, this data provides one sample per day
    • sunmars_km: the distance in kilometers between the Sun and Mars
    • earthmars_km: the distance in kilometers between the Earth and Mars
    • sunmarsearthangle_deg: Sun-Mars-Earth angle in degrees
    • solarconstantmars: solar constant at Mars in \(W/m^2\)
    • eclipseduration_min: total durations of all eclipses in the day, in minutes
    • occultationduration_min: total durations of all occultations in the day, in minutes

Observation data

Observation files are also in CSV format and follow the nomenclature:

  • power--YYYY-MM-DD__YYYY-MM-DD.csv
    • YYYY-MM-DD__YYYY-MM-DD indicates the start and end date in Earth UTC time.

These files are present only in the training set (train_set/) as they contain the measured electric current of 33 thermal power lines. Each file contains 34 columns:

  • ut_ms: unix timestamp in milliseconds
  • NPWD----: 33 columns/parameters with the electric current measurements

Usually one observation is made every 30 seconds or 60 seconds. The goal of the competition is to predict the average electric current per hour.


Evaluation

The goal of this competition is to predict the average current of 33 thermal power lines per hour of the Mars Express mission from 2015-01-01 to 2017-03-01.


Prediction file

For the prediction, we ask you to provide the average current (NPWD) in each of the 33 thermal power lines during the fourth year. The format of your submitted file must be a CSV file with the following columns:

  • ut_ms: unix timestamp in milliseconds
  • NPWD----: 33 columns/parameters with the predicted average electric current measurements

The data download contains a sample prediction file named power-prediction-sample-2015-01-01_2017-03-01.csv which shows the format.

Note that for a valid submission, the file must contain 34 columns and (TBD) lines of data plus one header line. The timestamps and column headers must be the same as in the sample prediction file. The submission system will inform you of any problems in your submission file (size, Inf, NaN...) as it will not accept any invalid file.


Evaluation

Each submission file will be evaluated against the fourth Martian year's actual average electric current data using the Root Mean Square Error (RMSE) measure.

$$\epsilon = \sqrt{\frac{1}{N M} \sum{(c_{ij} - r_{ij})^2}}$$

  • \(\epsilon\): root mean square error
  • \(c_{ij}\): predicted value for the ith timestep in the test set of thejth parameter
  • \(r_{ij}\): reference value for the ith timestep in the test set of the jth parameter
  • \(N\): the total number of evaluated measurements \(i \in [1,N]\) with \(N <= 16488\)
  • \(M\): the number of parameters \(j \in [1,M]\) with \(M = 33\)

For the public leaderboard your data will be tested only on a portion of the test data. The final leaderboard may thus vary.


Some solutions from the previous Mars Express Challenge

Stephanos Stepahni - ranked 2nd https://github.com/stephanos-stephani/MarsExpressChallenge

Fornax - ranked 3rd https://github.com/dataadventures/mars_express_orbiter

Alex Bauer - ranked 4th https://github.com/alex-bauer/kelvin-power-challenge

Rules

Next to the Code of Honor the following rules apply to the competition.


Eligibility

  • The Mars Express Power Challenge is open to everybody, including space experts, data-mining experts and citizen scientists.
  • No nationality restrictions apply for the submission of solution files.

Important Dates

  • June 19, 2017: Challenge starts
  • July 24, 2017: Submissions deadline;
  • July 26, 2017: Final leaderboard is published

Winner

  • The winner has been declared (see rules below) he/she must submit the code and a pdf document outlining, briefly, the method used.
  • Should the winner be unavailable to comply with the above rules, the second ranked team will be contacted instead, and so on.
  • All participants have the chance to submit their code and a pdf document describing their solution to the attention of the Mars Express Orbiter operators. Received solutions will have a chance to be incorporated in the operation of the spacecraft, in which case authors will be kept "in the loop" at every stage.
  • The winner is determined by the lowest score after the deadline when all data has been evaluated.
  • In case of a tie, the earlier submission date wins.
  • The final decision on the winner is subject to whether the participants conformed to the rules as checked by ESA Advanced Concepts Team.

Prizes

  • No, there is no money for the winner, "only" the chance to improve the operations of the Mars Express spacecraft -> thus extending its life -> thus increasing the scientific data that this amazing mission is producing -> thus advancing humankind's knowledge.
  • Both the winner and the runner-up teams will be awarderd one free registration each for the upcomming ECML PKDD 2017 (European Conference on Machine Learning & Principles and Practice of Knowledge Discovery), and given the opportunity to present the method and discuss its implications and possible improvements with the Mars Express operators.
  • All the interested participants will have the chance attend the conference and be part of the disusssions regarding the methods, the problems and the soultions of the challenge.

License

By submitting their software, the authors grant to the organizers a license to use it for the purpose of verifying the challenge results and carrying out post-challenge analyses on other data. The authors retain all rights to their software. ESA will keep the software confidential. ESA Open Access encourages you to publicly release your model code and documentation under one of the Open Source Initiative-approved licenses listed under the heading "Popular Licences" in the Open Source Initiative's list of approved licenses at http://opensource.org/licences. Data generated from the ESA data is to be released under creative commons licenses. All datasets provided are CC BY-SA 3.0 IGO.


Submission

  • The use of additional publicly available data is allowed when its license allow this use.
  • Before the competition deadline, solutions, notebooks, code snippets are encouraged to be shared publicly. Previous material can already be found on different code sharing platforms.
  • The submission limit is three submission within 24 hours per team.
  • The leaderboard during the competition uses only a part of the submission data and might not reflect the final ranking.

Teams

  • Every team is only allowed to submit with one account.
  • Merging teams is not allowed.