{ "id": 5957, "title": "Hi guys, need some clarifications", "url": "/competitions/flight2-milestone/discussion/5957", "author_name": "", "post_date": "2013-10-04T03:35:45.710Z", "votes": 1, "comment_count": 5, "views": 2891, "content": "

I'm new to this quest, so I also want to get this straight.

\n

What I know as a fact:

\n

1. The optimization objective, namely how the total cost is exactly calculated, is detailed in the source code of simulator.

\n

2. The routing solution is applied to each flight individually while others are restored to their actual routes recorded on that day. This means we cannot globally coordinate flights (which is odd for real world application).

\n

3. No information after cut-off time is allowed. There're 2 parts we need to predict: the wind direction, velocity etc. along the route of a flight; and airport conditions when the flight reaches destination. The later contains airport weather condition like wind, dewpoint, temperature etc, and also the no. of arrival and take-off airplanes within a short period after the flight reaches the airport.

\n

 

\n

What I'm not sure:

\n

1. For the weather prediction part, a bunch of METAR, TAF, AirSigmet, FBwinds reports gathered days before the test flights(prior to 2013/07/05 and provided in FlightQuest2DataRelease1_AggregateTrain) are available to train the model. These reports, as well as the wind conditions in various location and atitude(Weatherdata-partial) before the cut-off time of the test flights(FlightQuest2DataRelease1_AggregateTestFeatures) are also provided. We can use these data to predict the wind conditions along the route and airport conditions when flight arrives.

\n

2. What about NOAA RAP? Is it allowed to use any data for any day except those after the cut-off time of the test flight days?

", "messages": [ { "id": "32014", "postDate": "10/04/2013 03:35:45", "content": "

I'm new to this quest, so I also want to get this straight.

\n

What I know as a fact:

\n

1. The optimization objective, namely how the total cost is exactly calculated, is detailed in the source code of simulator.

\n

2. The routing solution is applied to each flight individually while others are restored to their actual routes recorded on that day. This means we cannot globally coordinate flights (which is odd for real world application).

\n

3. No information after cut-off time is allowed. There're 2 parts we need to predict: the wind direction, velocity etc. along the route of a flight; and airport conditions when the flight reaches destination. The later contains airport weather condition like wind, dewpoint, temperature etc, and also the no. of arrival and take-off airplanes within a short period after the flight reaches the airport.

\n

 

\n

What I'm not sure:

\n

1. For the weather prediction part, a bunch of METAR, TAF, AirSigmet, FBwinds reports gathered days before the test flights(prior to 2013/07/05 and provided in FlightQuest2DataRelease1_AggregateTrain) are available to train the model. These reports, as well as the wind conditions in various location and atitude(Weatherdata-partial) before the cut-off time of the test flights(FlightQuest2DataRelease1_AggregateTestFeatures) are also provided. We can use these data to predict the wind conditions along the route and airport conditions when flight arrives.

\n

2. What about NOAA RAP? Is it allowed to use any data for any day except those after the cut-off time of the test flight days?

", "rawMarkdown": "", "votes": null }, { "id": "32024", "postDate": "10/04/2013 13:25:59", "content": "

I asked a similar question right before you. Please keep an eye there for answers too.

", "rawMarkdown": "", "votes": null }, { "id": "32033", "postDate": "10/04/2013 17:25:03", "content": "

[quote=Edward He;32014]

\n

1. For the weather prediction part, a bunch of METAR, TAF, AirSigmet, FBwinds reports gathered days before the test flights(prior to 2013/07/05 and provided in FlightQuest2DataRelease1_AggregateTrain) are available to train the model. These reports, as well as the wind conditions in various location and atitude(Weatherdata-partial) before the cut-off time of the test flights(FlightQuest2DataRelease1_AggregateTestFeatures) are also provided. We can use these data to predict the wind conditions along the route and airport conditions when flight arrives.

\n

[/quote]

\n

Yes, that is correct. 

\n

[quote=Edward He;32014]

\n

2. What about NOAA RAP? Is it allowed to use any data for any day except those after the cut-off time of the test flight days?

\n

[/quote]

\n

Future information is not allowed, so if that information is available after the cutoff time, then you cannot use it. Note that this is different from information about the future being known at the present time. Here's a concrete example. Assuming an 8:30am cutoff time, you know at 8am that a flight is scheduled to land at 3:15pm. It's information about the future, but you know it now (at 8am). Future information would be knowing that at 8:33am the arrival time is going to be updated to 3:18pm.

\n

I hope this clarifies things and explains the rule of thumb on navigating the cutoff time.

", "rawMarkdown": "", "votes": null }, { "id": "32081", "postDate": "10/07/2013 12:03:44", "content": "

[quote=Edward He;32014]

\n

What I know as a fact:

\n

...

\n

2. The routing solution is applied to each flight individually while others are restored to their actual routes recorded on that day. This means we cannot globally coordinate flights (which is odd for real world application).

\n

[/quote]

\n

Hi Edward,

\n

I don't think this is true. (If I'm wrong I hope an admin will correct me.) You actually can coordinate all the flights which are in the air at a specific cutoff time simultaneously. This is less exciting then you might think since the planes have no influence on each other. They could be at the same time in the same place and after that land without a scratch.

\n

 

\n

 

", "rawMarkdown": "", "votes": null }, { "id": "32091", "postDate": "10/07/2013 16:04:54", "content": "

How flights interact with each are described on the Basic Structure page, point 2.

", "rawMarkdown": "", "votes": null }, { "id": "32093", "postDate": "10/07/2013 17:05:46", "content": "

[quote=Jules;32081]

\n

 

\n

I don't think this is true. (If I'm wrong I hope an admin will correct me.) You actually can coordinate all the flights which are in the air at a specific cutoff time simultaneously. This is less exciting then you might think since the planes have no influence on each other. They could be at the same time in the same place and after that land without a scratch.

\n

 

\n

 

\n

[/quote]

\n

 

\n

What I mean is, you cannot simultaneously instruct each flight to get a better performance. For example suppose during a busy hour of an airport a bunch of flights are arriving at the same time, this can cost you a lot of hold time on air. What you can do is to ask a few flights to fly faster, even though it may increase the fuel burn of these flights, it might reduce the total costs of all.

\n

But it's just a thought. There must be practical reasons this competition doesn't do this: too hard to record and model all flights simultaneously for one. From what I see, this competition stresses more on the data prediction part than the routing optimization part. If you can predict the weather and ground condition accurately, routing a single flight isn't that hard.

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 32024, "author_name": "davidwilson0", "author_url": "", "post_date": "10/04/2013 13:25:59", "content": "

I asked a similar question right before you. Please keep an eye there for answers too.

", "votes": null, "replies": [] }, { "id": 32033, "author_name": "joycenv", "author_url": "", "post_date": "10/04/2013 17:25:03", "content": "

[quote=Edward He;32014]

\n

1. For the weather prediction part, a bunch of METAR, TAF, AirSigmet, FBwinds reports gathered days before the test flights(prior to 2013/07/05 and provided in FlightQuest2DataRelease1_AggregateTrain) are available to train the model. These reports, as well as the wind conditions in various location and atitude(Weatherdata-partial) before the cut-off time of the test flights(FlightQuest2DataRelease1_AggregateTestFeatures) are also provided. We can use these data to predict the wind conditions along the route and airport conditions when flight arrives.

\n

[/quote]

\n

Yes, that is correct. 

\n

[quote=Edward He;32014]

\n

2. What about NOAA RAP? Is it allowed to use any data for any day except those after the cut-off time of the test flight days?

\n

[/quote]

\n

Future information is not allowed, so if that information is available after the cutoff time, then you cannot use it. Note that this is different from information about the future being known at the present time. Here's a concrete example. Assuming an 8:30am cutoff time, you know at 8am that a flight is scheduled to land at 3:15pm. It's information about the future, but you know it now (at 8am). Future information would be knowing that at 8:33am the arrival time is going to be updated to 3:18pm.

\n

I hope this clarifies things and explains the rule of thumb on navigating the cutoff time.

", "votes": null, "replies": [] }, { "id": 32081, "author_name": "julesvanligtenberg", "author_url": "", "post_date": "10/07/2013 12:03:44", "content": "

[quote=Edward He;32014]

\n

What I know as a fact:

\n

...

\n

2. The routing solution is applied to each flight individually while others are restored to their actual routes recorded on that day. This means we cannot globally coordinate flights (which is odd for real world application).

\n

[/quote]

\n

Hi Edward,

\n

I don't think this is true. (If I'm wrong I hope an admin will correct me.) You actually can coordinate all the flights which are in the air at a specific cutoff time simultaneously. This is less exciting then you might think since the planes have no influence on each other. They could be at the same time in the same place and after that land without a scratch.

\n

 

\n

 

", "votes": null, "replies": [] }, { "id": 32091, "author_name": "joycenv", "author_url": "", "post_date": "10/07/2013 16:04:54", "content": "

How flights interact with each are described on the Basic Structure page, point 2.

", "votes": null, "replies": [] }, { "id": 32093, "author_name": "edwardhe0", "author_url": "", "post_date": "10/07/2013 17:05:46", "content": "

[quote=Jules;32081]

\n

 

\n

I don't think this is true. (If I'm wrong I hope an admin will correct me.) You actually can coordinate all the flights which are in the air at a specific cutoff time simultaneously. This is less exciting then you might think since the planes have no influence on each other. They could be at the same time in the same place and after that land without a scratch.

\n

 

\n

 

\n

[/quote]

\n

 

\n

What I mean is, you cannot simultaneously instruct each flight to get a better performance. For example suppose during a busy hour of an airport a bunch of flights are arriving at the same time, this can cost you a lot of hold time on air. What you can do is to ask a few flights to fly faster, even though it may increase the fuel burn of these flights, it might reduce the total costs of all.

\n

But it's just a thought. There must be practical reasons this competition doesn't do this: too hard to record and model all flights simultaneously for one. From what I see, this competition stresses more on the data prediction part than the routing optimization part. If you can predict the weather and ground condition accurately, routing a single flight isn't that hard.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "32014": "", "32024": "", "32033": "", "32081": "", "32091": "", "32093": "" }, "source": "meta" }