Google uses cookies and data to:
- Deliver and maintain services, like tracking outages and protecting against spam, fraud, and abuse
- Measure audience engagement and site statistics to understand how our services are used
If you agree, we’ll also use cookies and data to:- Improve the quality of our services and develop new ones
- Deliver and measure the effectiveness of ads
- Show personalized content, depending on your settings
- Show personalized or generic ads, depending on your settings, on Google and across the web
Sky Light Sensor is useful to monitor in-block and off-block time of flights. Sky Light Aircraft System is a way to effective airport operation management. Definition The estimated time at which the aircraft will commence movement associated with departure.
For non-personalized content and ads, what you see may be influenced by things like the content you’re currently viewing and your location (ad serving is based on general location). Personalized content and ads can be based on those things and your activity like Google searches and videos you watch on YouTube. Personalized content and ads include things like more relevant results and recommendations, a customized YouTube homepage, and ads that are tailored to your interests.Click “Customize” to review options, including controls to reject the use of cookies for personalization and information about browser-level controls to reject some or all cookies for other uses. You can also visit g.co/privacytools anytime.
Airports are busy places where different stakeholders have key roles and a common goal to manage safely and efficiently the flow of flights departing and arriving. However, airport infrastructures are not exploited in the most optimal manner and increasing traffic makes it difficult for operations to be proactive rather than reactive. This is due to the lack of good information sharing procedures, each of the stakeholders involved in operations has a piece of the information rather than the global picture.
With a 3 hour look-ahead, the airport collaborative decision making (A-CDM) process enables all stakeholders to benefit from sharing the same information as early as possible in order to take informed decisions. Milestones and relevant flight details are updated and shared in order to have accurate Off-Block Times (OBT) and Take-Off Times (TOT) for better situational awareness. This information sharing leads to better traffic flow management at the network level.
In this article, we will explain what is the Target Off-Block Time (TOBT) and how it is used in the A-CDM concept. We will also focus on why its accurate estimation is of great value for the airport operations. After that we will present the current capacity of Innov’ATM Airport dedicated product AirportKeeper in predicting TOBT in order to prevent disruption due to the turnaround process management. Finally we will explain how we explored the use of Machine Learning algorithm in order to enhance the TOBT prediction and what are the results on the prediction accuracy.
2.1 – What is the TOBT used for
TOBT is the target time set as off-block departure time by the airline or ground handler in charge. At this time:
airplane’s doors are closed
boarding bridge is removed
push back vehicle is available
airplane is ready to start up/push back upon clearance.
TOBT and its updates improve predictability during the turnaroud process of aircraft.
It contributes to better allocation of parking stand ressources through knowledge of parking stand occupancy time and aircraft stand allocation accuracy.
It contributes to better workload management of ATC controllers that can anticipate the number of flights that will use the platform simultaneously.
By using Variable Taxi Times, a proper prediction of take-off times can then be communicated to the Network Manager Operation Center as an input for the management of European network.
This TOBT is the base time used by the Air Traffic Controllers to organize the sequence of aircraft to take-off and deliver push back clearance as described in paragraph 1.1.1 & 1.1.2 of our previous article Reducing operational costs with Innov’ATM / AI – Episode 1: Ground holding time… This is why its accuracy and stability is important in order to provide an accurate and stable flight departure sequence.
2.2 – What is the TOBT life-cycle
The TOBT is declared manually by the airline or the ground handler in charge. It is well observed that the TOBT is updated very lately in the TOBT lifecycle. This is due to various reasons among which:
Ground handlers are busy trying to get the A/C ready for push back on schedule.
Parking fees are proportional to stand occupation time, so ground handlers job is to try to minimize this time (it has to be noted that ground handlers get penalties when the delay can be blamed on them).
Ground handlers do not want to loose their place in the departure sequence, which would result in a bigger delay (when TOBT changes, some rules are applied by ATC that may result in removing the flight from the departure sequence and rescheduling it later)
Some events are independent from ground handlers will, and they have no control on it (boarding longer than usual, passenger no show, …)
As a consequence, TOBT are updated only when the problem is confirmed, and when no other option is possible, which means updates happen very lately in the turnaround lifecycle. This is a pain point in A-CDM Information Sharing objective. It then becomes crucial to be able to predict the TOBT that will be issued in order to increase Information Sharing among all stakeholders and optimize global efficiency. We will refer to the prediction of target off-block time as the POBT in this article.
The TOBT prediction (i.e POBT) allows operators to identify potential disruptions between what is currently the shared target of off-block times and the actual capacity of the aircraft to be ready for off-block regarding its current status. Today Innov’ATM AirportKeeper product already integrate a POBT computation which is displayed side to the TOBT value in its information sharing flight list.
This POBT is for example used to raise an alert indicating that the TOBT shall be updated by the ground handler when POBT > TOBT + 5 min.
The current POBT computation algorithm we use in operation is a simple and logical business rule based algorithm that relies on flight informations available along the life cycle of a turnaround flight. We will name the POBT computed by this algorithm POBT(R) in the rest of this article.
Based on the current status of the flight, this algorithm combines the most accurate informations in the following values:
SOBT – The scheduled off-block time of the flight.
EOBT – The estimated off-block time of the flight.
CTOT – The calculated take of time of the flight.
O-ATOT – The actual take of time of the flight at its origin airport.
ELDT – The estimated lading time of the flight.
EXIT – The estimated taxi-in time of the flight.
MTT – The minimum turnaround time of the flight.
EXOT – The estimated taxi-out time of the flight.
ALDT – The actual landing time of the flight.
AIBT – The actual in-block time of the flight.
As an example, as soon as the aircraft land at the airport, the POBT(R) is computed by combining the Actual Landing Time (ALDT) with the Estimated Taxi In Block Time (EXIT) and the Minimum Turnaround Time (MTT) of the flight:
The current accuracy of POBT(R) compared to the final value of TOBT has been evaluated on a month of data at one of the airport where AirportKeeper is deployed. The evaluation has been performed in regards to the prediction horizon, i.e. the time before flight scheduled departure (SOBT) at which we would like to predict the final value of TOBT. The prediction horizons used are 3 hours, 2 hours, 1 hour, 40 mins, 30 mins, 20 mins and 10 mins before SOBT.
First of all the distribution of the average error shows that most of the predictions are within a +/-5 minutes ranges.
We then computed the mean absolute error (MAE) of POBT(R) vs the final TOBT. The MAE is an arithmetic average of the absolute difference between predicted and actual values.
It is clearly seen how the prediction error decreases with the decrease of prediction horizon, from 11 minutes 3h before the flight departure to 8.5 minutes 10 minutes before the flight departure.
Even if this prediction is already usable in operation, we decided to give a try to Machine Learning (ML) algorithm to see if we could get a more precise prediction for POBT computation.
Artificial Intelligence (AI) is the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings. The aim of AI is to improve computer functions which are related to human knowledge, for example, reasoning, learning, and problem-solving. ML is a branch of artificial intelligence that can be defined as the study of computer algorithms that allow computer programs to automatically improve through experience.
If you are interested in getting more information on the concept of ML and how it is applied, you can consult paragraph 2.1 and 2.2 our previous article Reducing operational costs with Innov’ATM / AI – Episode 1: Ground holding time.
The POBT computed thanks to ML will be named POBT(ML) in the rest of this article.
In the ideal word, the best situation that could occur is when all flights respect their schedule precisely, i.e. TOBT= SOBT for all flights. In reality that is evidently not possible, and delay may be introduced to the schedule. This delay is computed as :
Our aim is to predict more accurately this TOBT_delay, i.e. predict for how much a flight would divert from its schedule off-block time. Thanks to this TOBT_delay prediction we will be able to predict the TOBT related to the flight SOBT, and thus compute a POBT as :
After validation and cleaning of the historical data available to train our ML model, the result flight set contains 8249 flights.
Following features derived from flight data were used:
day: day of week (derived from SOBT)
hour: hour of day (derived from SOBT)
nbScheduled: number of flights scheduled to depart within the hour of SOBT
acTypeIcao: aircraft type
airlineIcao: operating airline
departureParking: flight parking
adesGroup: destination zone (retrieved from first letter of ADES code)
EOBTdiff: estimated delay at prediction horizon, i.e. difference between last EOBT received by the prediction time and SOBT, in minutes
TOBTdiff: expected delay at prediction horizon, i.e. difference between last TOBT received by the prediction time and SOBT, in minutes (see to Figure below)
numTOBT: number of TOBT updates received up to the prediction horizon
hasPrevious: boolean, 1 if flight has preceding arrival flight, 0 otherwise
arrivalMilestone: milestone of preceding arrival flight (if any) at prediction horizon
adepGroup: departure zone of preceding arrival flight (retrieved from first letter of ADEP of preceding arrival flight if any)
SIBTdiff: difference between SIBT of preceding arrival flight (if any) and SOBT, in minutes
arrDelay: expected or actual delay of preceding arrival flight (if any) at prediction horizon, i.e. difference between most relevant IBT of preceding arrival flight (in order, AIBT, TIBT, EIBT) received by the prediction time and its SIBT, in minutes.
Of Block Scratch
As for the POBT(R) algorithm, we evaluated the distribution of prediction errors at different prediction horizons (3h to 10 min before SOBT).
Off Block Utc
As can be seen, the distribution is well-centered at 0 and the majority of flights have the prediction error in the range of +/-5 minutes. It is also seen that the number of flights having smaller errors is increased with the decrease of prediction horizon.
We then computed the MAE for different horizons, and compared the results with the previous POBT(R) algorithm.
The POBT(ML) is slightly more accurate than the POBT(R) that is currently used in our AirportKeeper product.
POBT computation by ML algorithm (i.e POBT(ML)) provides very good results. Using features mostly related to flight data, we were able to enhance significatively the accuracy of prediction in comparison to the Off-Block Time computed based on business logic (i.e POBT(R)).
Thus, the ML-based Off-Block Time computation is an option integrable in our AirportKeeper product. When in operation, more data related to actual airport (meteo, pax count, boarding progress status…) will allow to train more and more the algorithm and provide a very good level of Off-Block Time prediction.