Already developed at the strategic level, time windows are assessed in the tactical phase also from a flight performance point of view, i.e., fuel consumption and arrival flight delay. We consider a time horizon from a few hours before flight departure up to 20 minutes before arrival, when the horizon of the Arrival Manager process begins. The input of this assessment are the time windows and routes obtained at the strategic level. Here, uncertainties that at the strategic level are difficult to capture, such as weather forecast errors for the day of the operation or departure delays, are considered (the details are a matter of a specific tactical scenario to be defined at the beginning of the project).
Adjusting Time Windows
Both the strategic time windows and these uncertainties are fed into a stochastic dynamic programming framework to decide whether the size of the time windows needs to be further changed/adapted such that aircraft fuel consumption and arrival delays are minimized. As a result, the strategic Time windows are potentially updated en-route, i.e., their starting and closing times are periodically re-evaluated, having as objective the minimization of expected costs with respect to fuel consumption and arrival delays.
Flexibility for airspace users
Next, the trajectories obtained in this process are assigned a measure of flexibility, that we term time windows (Time windows). The time windows measure the flexibility of a flight in terms of time slack available to a flight, within which no capacity problems are expected to be created along the agreed trajectory. Apart from Time windows assigned to flights, the critical elements of the planned network configuration (i.e. sectors or airports) are also identified. Thus, already at the strategic level, an indication of the flexibility or criticality of flights and critical network elements can be obtained.
The ADAPT strategic approach will be then evaluated against the tactical horizon both at a network level and from a flight centric point of view.
Simulating Stochastic Environment
We simulate the stochastic environment (weather forecasts, departure delays) using the open-source BlueSky simulator developed at TU Delft. In contrast to the ELSA model, BlueSky has embedded functionalities for aircraft performance modeling using BADA 3 data, and can simulate the arrival phase of flights. Starting with the strategic time windows as initial condition, potential changes in these time windows are generated, as indicated by the stochastic dynamic programming framework mentioned above. The aim is to assess whether changes in the initial time windows are frequently needed to minimize the expected flight cost in terms of fuel consumption and arrival delays. The output of the assessment are potential en-route deviations from the strategic time windows. This provides an overview of the impact of time windows usage not only at the level of sector demand-capacity balancing, but also from an aircraft performance point of view.
Building on BlueSky simulator
In contrast to the time windows at the strategic level, the flight-centric tactical assessment of time windows requires an added aircraft functionality that can handle time windows on the flightdeck, keeping track of the impact of time windows on fuel usage and flight time. We therefore introduce a new functionality to the aircraft performance model available already in BlueSky, to allow for optimal aircraft operation within a given time window. Whenever the estimated time of arrival (ETA) of a flight, which is already en-route, stays within the specified time window, the aircraft operates at its optimal speed. Only when the flight is expected to operate outside the time window (i.e., ETA outside the time window), the aircraft speed regime is adjusted accordingly at the cost of some additional fuel. This new functionality will be added to the BlueSky aircraft performance model to be able to assess the concept of flight-centric Time windows at the tactical level.
Addressing scaling problems
For large systems, the proposed method becomes computationally intensive. To address this, we will scale up the stochastic dynamic method from flights arriving to a single arrival sector from any European departure location, to multiple arrival sectors. The time span of the analysis is up to 10 hours of operations in a day. We will introduce an approximate dynamic programming extension that allows scaling up to the arrival sectors of all major European airports