Predicting the amount of adhering snow on bogie part of Hokuriku Shinkansen, High speed rail
- 2018/01/31 (Closed)
- Number of Participants
Gold Prize : ¥1,000,000
Silver Prize : ¥700,000
Bronze Prize : ¥300,000
Extra Price: Special attendance for related facilities with West Japan Railway Company
- Host / Client
Tuesday, Dec 5, 2017 - Replaced an image of evaluation metric with correct one.
OverviewWhen high speed rail rolling stock, Shinkansen, travels during snowfall, snow adheres on its bogie part. An increase in the amount of adhering snow may cause its falling into the track. This falling has a risk of destruction of track infrastructure and so on. Therefore, prediction of the largest amount of adhering snow on the bogie with the risk of falling on the day is required in order to make a decision on the necessity of the next day’s removing adhering snow on the previous day. For this reason, improving the prediction accuracy of the amount of adhering snow is expected super effective to support deciding on the necessity of removing snow on the next day. In this competition, participants will create a model that predicts the amount of adhering quantitatively by analyzing historical data of weather conditions, traveling conditions and amount of adhering snow. Though it may be a very complicated problem as it requires viewpoints of both meteorology and railway operation, we would like to make use of your great ideas, technical skills, and creativity to find an innovative solution and keep on improving the quality of railway operations.
Glossary of technical terms
Adhering snow: Amount of adhering snow on the rolling stock bogie. Kilometrage: Accumulated distance from Kanazawa station, which indicates a location of station, tunnel, and each measuring instrument. Train ID: An identifier given to each train in a diagram.
Operating section: Kanazawa station -> Itoigawa station Target variable: Amount of adhering snow on bogie part at the arrival of Toyama station for all railways during test set period Train set period: Jan 19 ~ Dec 31, 2016 Test set period: Jan 1 ~ Mar 31, 2017
One account per individual You may have one account per participant. No private sharing Privately sharing code or data is not permitted. Restriction on creating a model Creating a predictive model by plagiarism or plagiarism of an existing research is not permitted. External data use You may use open data other than competition data as long as they are free. Software You must use only open source software for creating a model (i.e. python, R) Model should not lack consistency Model should not require any additional costs in general usage, and it needs to be confirmed to reproduce their predictions and be continuously used. Model must use the same logic to generate predictions for all the given samples. Manual labeling on predictions is not allowed. (If you have any enquiries about the criteria, please contact us.) Do not use information or data that would not have been known or available Following data can be used on target date of prediction. ①From Meteorological Agency database, participants can use hourly information of temperature, precipitation, wind speed, relative humidity, and weather during the same day as the target date of prediction. These information are provided by Japan Weather Association and are forecasted the day before. Let's say we are making a prediction on Jan 11, 2017: data with index before 2017-01-12 00:00:00 can be used. Please note: 24:00 of one day is the same time as 00:00 of the following day. ② From Meteorological Agency database, data of snow depth, anemometer and the other items not listed in the above ① can be used as long as they are disclosed on or before 4 AM on the target date of prediction. Participants can also use external data as input data as long as they can be acquired on or before 4 AM on the target date of prediction. Let's say we are making a prediction on Jan 11, 2017: data with index before 2017-01-11 04:00:00 can be used. ③ Please not that the weather information (other than ①) considered necessary for improving prediction accuracy of the amount of snow accretion can be used as secondary input data by estimating the relationship with (primary) input data of ① (See figure below).
Friday, Dec 1, 2017 - Competition Launch Date Wednesday, Jan 31, 2018 - Prediction Submission Deadline Monday, Feb 5, 2018 - Source Code Submission Deadline (※Only for winning candidates) Mid-February (To be determined) - Determination of final rankings
Tips for modeling
Though participants can only use temperature, precipitation, wind speed, relative humidity, and weather as input data for prediction dates in this competition, we would like to introduce the ideas to estimate other weather conditions from the meteorological point of view based on these meteorological conditions (*For more details, please refer to [doc_2_en.pdf] on the data download page). A method to estimate amount of snowfall from precipitation Precipitation during snowfall indicates the depth of water when the drifted snow melts into water. Amount of snowfall indicates the depth of snow (snow depth) that has been precipitated. Even with the same precipitation, the value of snowfall will change depending on the density ratio of water and snow. Since it is shown that the density of snow changes depending on the temperature from past experimental law, it is possible to estimate the amount of snowfall from the temperature and the precipitation during snowfall. A method to estimate solar radiation from sunshine duration The amount of solar radiation is an indicator of the strength of energy due to the sun's sunshine, and an increase in the amount of solar radiation during snowfall or snowfall acts in the direction of melting snow (the snow density increases). It is possible to estimate based on the relational expression between sunshine hours and solar radiation at a point where solar radiation data is not observed. As will be described later, it is possible to estimate the sunshine hours from the weather. A method to estimate sunshine duration from the weather It is possible to find a correlation between weather and sunshine duration by analyzing statistically based on the historical weather data. For reference, we show the relationship between weather and sunshine duration in case of using the Meteorological Agency data of Kanazawa city which is prepared in this competition. By using the above relation, it can be used to estimate the sunshine duration condition at the point where the sunshine duration is not observed and the target date of prediction.
・Historical weather information: Meteorological Agency ・Weather forecast: The meteorological society of japan (tenki.jp) ・Method of predicting the amount of adhering snow on rolling stock bogie ・Snow control measures on the Tohoku Shinkansen line between Hachinohe and Shin-Aomori: Part 6 ・Science of the atmosphere close to the ground surface
- Number of submissions：
- Number of participants：
|Rank||User name||Score||Number of submissions||Last|
|19||Hiroki Umehara||0.00300||45||2017/12/15 00:30|
|48||Kyo Naganuma||0.00333||3||2017/12/28 14:48|