A data - driven - probabilistic queuing network model for european air traffic flow

It is widely understood that congestion and delay in transport networks can spread from one part of the network to another. In the case of airport, this means that airport congestion in one region can cause delays at all network. As air travel demand increases over years and some airports operate close to their capacity limits, it is necessary and important to study the relation between capacity and demand of any given airport, which causes delay on the airports specially when we are dealing with large networks of airports to predict and balance their demand and capacity to decrease the amount of the delay. In this thesis we will present a model, Airport Network Delay model (AND model) that is able to estimate delays and propagated delays to the whole airports network. Actually from one perspective fundamental of some ideas of AND model is similar to Long et all LMINET model. Both of them are analytical models but AND model has more advantages than LMINET which does not use aircraft itineraries. This thesis closely followed AND model (Prygiotis PhD thesis) but we implemented simulation method instead of analytical methods. After deep analysing and comparing the results of these two different analytical and simulation approach, we realize our approach is faster than original AND model's procedure. This model is formed by two fundamental parts, Queuing Engine (QE) and Delay Propagation Algorithm (DPA). First part is used for estimating the delays occur at the individual airports as node of the network by utilization of the queuing theory. Queuing Engine investigate the scheduled demands and capacity of the airports individually to find the delays take place in nodes of the network due to a day by simulating the M(t)/Ek(t)/1 queuing system. Second important part of this model means Delay Propagation Algorithm, estimate the delays that take place due to the itineraries of the aircrafts and propagate such sufficient delays to the network of airports. 24 hours of the day divided to the equal time steps named sub-period with 15 minutes length. DPA operating could be divided in to the delay estimator and delay propagator parts. First, the algorithm compute delays to decide whether estimated delay value is sufficient enough to propagate or not. Estimating expected delays is calculated refer to some nomenclatures and slack values in the ground turnaround time. Service times is calculated by analysing the real data and follow a time varying Erlang distribution. If expected delay is such efficient to propagate, reminder scheduled flights of the day will affect by this delay. This operation runs for each time steps by tracking the flights itineraries. Then it will update the scheduled arrival/ departure of the destination airport for rest of the time steps. So the most important and main input of this model is itineraries of all aircrafts. Actually designed model iterates between QE and DPA and each runway system is modelled as a single server.