TPI

Traffic prediction improvements

The Traffic predictions improvements (TPI) project at EUROCONTROL MUAC aims at improving predictability of traffic to optimise the usage of air traffic controllers (ATCOs) time and to reduce uncertainty of capacity predictions. 

Our goal is to enhance our trajectory predictions by anticipating operational behaviours in specific conditions (skipped sectors by air traffic control (ATC), high complexity hotspots) and through better use of historical data (e.g; estimated take-off time prediction, 4D trajectory, flown route to be expected).

Context and business driver 

The Integrated Flow Management Position (iFMP) at our Maastricht Upper Area Control (MUAC) Center  gives a view on upcoming air traffic controller workload in the different sectors.  iFMP allows to take informed decisions about the most optimal sectorisation, balancing traffic demand and available manpower, and to identify efficient traffic regulations. The tool combines NM traffic forecasts (ETFMS) with live data from MUAC operational ATC systems. Time horizon of iFMP is 30 minutes to 6 hours. 

iFMP retrieves airline preference via the ATM Portal (ATMP). It also interfaces with the Controller Working Positions to coordinate traffic measures on individual flights with the relevant air traffic controllers.  

In addition, iFMP hosts a cluster detection function to pinpoint traffic hotspots in the 10min-20min time horizon. Purpose of iFMP is to resolve traffic complexity proactively, i.e. before air traffic controllers are confronted with it. 

Prerequisites are reliable traffic predictions. Unfortunately, major uncertainties exists in the iFMP time horizon. TPI improves the quality of traffic predictions by using advanced AI algorithms and the wealth of data that MUAC has collected over the years. 

Functions

The AI algorithms are applied to several key functions in the iFMP.

Prediction of flight routes

The route flown by an aircraft often deviates substantially from the route in the flight plan due to instructions by air traffic controllers. The AI algorithm has learned to recognise patterns in historic data, and can predict the actual route prior to flight entry by considering flight data, status of military areas, and time during the day. (in use since 2018 for selected flows) 

4D trajectory prediction

4D flight trajectories calculated by the MUAC flight data server are improved in real-time by an AI algorithm that also considers recent radar data and environmental data. The AI algorithm produces a more accurate 4D prediction in the 5min-30min horizon. It has e.g. learned to recognise slow climbers and knows that those flights will enter the airspace at a different point/time than originally calculated.  

Prediction of Take-Off Times 

The AI algorithm corrects the take-off time estimate based on a wide set of predictors, including prior flights, regulations and traffic demand.

Prediction of sector skips 

The way air traffic controllers hand-over flights from one sector to the next often deviates from the geographic sector sequence. This pollutes sector workload forecasts. An AI algorithm has learned which sectors will be skipped by air traffic controllers. 

Prediction of complexity 

Hotspots of complex traffic are intrinsically not based on a fixed predicted route, but depends on the potential actions of a group of flights. To improve the identification of these hotspots in the 5-30 minutes horizon, a probabilistic forecast of the flight future positions is extracted by an AI algorithm that considers recent radar data, flight plan and environmental data (e.g. military). Based on the possible “action areas” of each flight, it is possible to determine potential future zones of interactions with assigned probability, helping ATCO take mitigation actions in advance. (in off-line validation phase, planned for late 2021).