Methodology
The stand assignment method presented in this paper is composed of the two-module approach and experiments in a simulation model. The two-module approach generates optimized stand allocations based on the target flight schedule, historical data about schedule disruptions for the previous period, and user- defined assignment policies and optimization goals. After that, the obtained allocations are estimated in the simulation model that allows evaluating the environmental footprint quality of stand assignments generated in the two-module approach under the stochasticity of a real-life airport system.
Algorithm Description
This section gives a short description of the two-module approach that generates optimized stand assignments. A more general description can be found at Bagamanova and Mujica Mota (2020). 517Bagamanova and Mujica Mota
The two-module approach is composed of two elements. Module I takes care of estimating probabilities of schedule deviations from the airport historical data. These probabilities are expressed in the form of Bayesian distributional models and describe a likelihood of certain levels of schedule deviations for various flight characteristics available in the historical data (e.g. such as airline name, scheduled time of arrival, and day of the week). By considering probable disruptions in the assignment planning, it is intended to reduce the idle time that aircraft might have to spend waiting for the planned stand availability and related emissions
Module II assigns the target flight schedule to the available stands, respecting user-defined assignment policy and restrictions, considering most probable or user-defined probability level schedule disruptions in the stand occupancy times. Then the generated assignment is optimized with a genetic algorithm according to user-specified optimization goals. The result of such optimization is not necessarily an optimal solution, however, randomness used in the genetic algorithm in the form of crossover and mutation operators allows us to obtain a good quality solution in a reasonable time (Bagamanova and Mujica Mota 2020). The resulting stand assignment considers the stochasticity in the form of stand occupancy times deviations generated from the schedule deviations distributional models.
Optimization Objective
To increase stand assignment efficiency and mitigate pollutant footprint, produced by aircraft movement on the ground and aircraft idle waiting for stand availability, the following bi-objective optimization goal function has been implemented in the optimization component of Module II of the two-module approach:
min(w 1 ∗ O taxi + w 2 ∗ O hold ) (1)
The objective function (1) consists of the following individual objectives:
- Minimize taxi distance to and from the parking positions and therefore the related emissions:
O taxi = dsched.taxi ⁄ Max dairport
- Minimize the number of aircraft waiting for stand availability and, therefore, the idle use of engines:
O hold = ∑fl. hold ⁄ ∑ fl.
Where:
- dsched.taxi ⁄– the average taxi distance to and from the stand in the allocated schedule;
- Max dairport the maximum possible taxi distance at the airport for considered runway configuration;
- ∑fl. hold – the number of aircraft that must wait for the stand availability;
- ⁄ ∑ fl – the total number of aircraft in the schedule to allocate;
- wn – priority weight for the corresponding objective. In the scope of this paper, all the weights areequal to 1 to obtain a stand assignment equally balanced for both considered objectives. Forpractical use, different stakeholders of the airport can decide the weights based on their preferences.
In the original implementation of the two-module approach by Bagamanova and Mujica Mota (2020),the optimization objective function in Module II also included maximization of the use of contact stands.This is a general preference for many airports as it allows to fully benefit from terminal building in termsof providing passenger experience and reduces the number of ground service vehicles moving on the apron.Yet, for the scope of this paper, such an objective was excluded as the primary goal is to generate a stand assignment with minimized emissions. Nevertheless, it might be interesting to investigate theenvironmental cost of prioritizing contact stand use in the optimization component in future work.
CASE STUDY: MEXICO CITY INTERNATIONAL AIRPORT
This section discusses the application of the two-module approach for encountering more environmentally efficient stand assignment policies for a case study airport.
General Information
Mexico City International Airport (IATA code: MEX) is the main airport in Mexico with approximately 450 thousand landings and take-offs annually. There are two terminal buildings, separated by two parallel runways. These runways are never operated simultaneously due to proximity to each other. Such layout restricts MEX capacity and since 2017 it has been officially limited to 61 movements per hour with a maximum of 40 landings (SCT 2017)
In the scope of this paper, it is considered that 26 airlines are operating in two terminals in MEX, performing both international and domestic flights. From the total 91 stands available at MEX, only 84 were considered in this paper, as the rest is not used for passenger flights. Hence, Terminal 1 is represented by 11 open stands and 33 contact stands, among which 16 stands are dedicated to domestic flights and 17 to international. Terminal 2 is represented by 17 open stands and 23 contact stands, where 13 are used for domestic flights, 10 – for international.
Schedule Disruptions and Emissions
On a global level, in 2018 Mexico generated approximately 1.5% of global air passenger transport-related emissions (Graver et al. 2019). MEX is located in the direct proximity of the urban zones of Mexico City, which makes the airport significantly affect air quality and noise levels of the city. According to SEDEMA (2018), MEX produces around 15% of the total pollutant emissions of Mexico city
In 2017 Mexico has officially joined a global initiative for carbon-neutral air transport operations (ICAO 2020), which implies that all country airports have to follow ICAO emission reduction policies and standards. Despite these facts, up to the date of writing this paper official MEX website did not publish any official estimations of airport emissions level nor disclosed any measures to reduce the environmental footprint of its operations
MEX frequently suffers from punctuality problems. In 2018 only 67% of all flights were performed on time (SCT 2019) with more than 20% of departing flights being delayed by 46 min on average (Flightstats 2018). Considering such a high level of perturbations and recent engagement in global pollutant footprint reduction initiative, MEX becomes a good target for application of the two-module approach to discover the hidden potential for emissions reduction related to stand assignment planning.
Implementation of the Two-Module Approach
As input data for this study, we used an official performance report for a period from 28.05.2018 to 03.06.2018, retrieved from International Airport of Mexico City (2018). This report consisted of more than 8,000 flights with actual and scheduled arrival times, flight numbers, airline names, and type of aircraft used. In the chosen week approximately 7% of arriving flights deviated for more than one hour from their schedule. More than 53% of scheduled arrivals suffered from a substantial delay of more than 15 min, which is a significant perturbation for a congested airport
Due to the unavailability of actual data on turnaround times and arrival-departure aircraft correspondence, it was assumed to use only arriving passenger flights from the obtained report and define 60 minutes turnaround time for all flights in the performed experiments. Such limitations reduced the number of flights to 3,914 arrivals, where 31.7% were international flights and 68.3% – domestic
The selected data of 3,914 flights have been processed in Module I and the Bayesian models for arriving time deviations were built, assuming the correlation of deviations with airline name and hour of scheduled arrival. The detailed description of the resulting parameters of regression models, composing the summative Bayesian model, and output of Module I can be found at Bagamanova and Mujica Mota (2020)
Lastly, Module II created an assignment, considering most probable scheduled deviations, assignment policy restrictions, and optimized it according to the objective function (1). As the two-module approach is considered to be a more effective replacement to traditionally used buffer times, for the generation of stand assignment in Module II no buffer times were intentionally added between consecutive flights assigned to the same stand. The resulting assignment statistics are shown in Figure 1
Every airport has its own stand assignment policy restrictions, which implies certain use of the stands. The following are the restrictions considered in the presented algorithm:
- Domestic and international flights must be assigned to the specific stands in the designated zones. These are internal specifications of the airport e.g. international flights are assigned to stands that have access to the designated border control areas;
- Flight delays must be considered in the assignment (according to conditional probability distributions from Module I). In this paper, only arrival delays are considered due to unavailability of ground handling data and correspondence of arriving aircraft to departing aircraft;
- An assigned stand must correspond to the size of an aircraft (large aircraft require extra space due to larger wingspan). This is implemented through the identification of allowed stands for each flight on the stage of processing the input data in Module II.
Figure 1: Assignment statistics for Module II generated stand allocation.
As can be observed from Figure 1, most of the flights were assigned to stands located not too far from the runways. In Terminal 1 approximately 61.1% of scheduled flights were assigned to a stand located closer than Terminal 1 average taxi distance of 4.2 km from the runway; for Terminal 2 – 61.3% of flights were assigned to the stands with less than average Terminal 2 taxi distance of 5.6 km. Naturally, some of the flights had to be assigned to further located stands due to assignment policy constraints, designated border control zones, and unavailability of closer located stands. Nevertheless, Figure 1 demonstrates the algorithm’s success with the minimization of taxi distance
One of the limitations of the data used for this study is the unavailability of actual historical MEX stand assignments. Therefore, for the moment, it is impossible to compare the quality of the two-module approach results with actual MEX stand assignments. Thus, to evaluate the quality of the obtained assignment and owing to the absence of actual historical stand assignments at MEX, the two-module approach assignment was tested in the environment of the MEX simulation model, as described in the next section. The detailed description and validation of this simulation model can be found at Mujica Mota and Flores (2019)
Simulation Experiments
The principal objective of using a simulation model in this study is to evaluate the effects of consideration of schedule deviations in the stand assignment on the taxi-related emissions in close-to-reality conditions and encounter ways to improve airport performance and emissions level. The simulation model used in this study allows us to incorporate stochastic elements (such as stop-go situations, waiting for push-back at the gate) that were not considered in the assignment generation, but do influence aircraft movements on the ground in the real life.
For each simulation replication the following performance indicators were tracked:
- total taxi distance for all aircraft of the allocated schedule: d total taxi = ∑ N
i=1 (d in i + d out i ) ;
- total taxi time for all aircraft of the allocated schedule: t total taxi = ∑ N
i=1 (t in i + t out i + t wait i ) ;
- total amount of taxi-related pollutant emissions e total taxi = t total taxi ∗ F NO + t total taxi ∗ F CO ;
where:
- d in i – distance traveled by aircraft i from runway exit to a stand;
- d out i – distance traveled by aircraft i from a stand to runway entry point;
- t in i – time traveled by aircraft i from runway exit to a stand;
- t out i – time traveled by aircraft i from a stand to runway entry point;
- t wait i – time spent by aircraft i waiting for stand availability;
- F NO and F CO – emission factors for NOx and CO 2 respectively;
- i… N – number of aircraft.
Emission factors depend on the engine characteristics, type of fuel used, and aircraft weight among others (ICAO 2019b). Due to the unavailability of any actual data about engine specifications and aircraft weight for the studied flight schedule, the amount of total emissions e total taxi was calculated assuming constant taxi speed and the taxi emissions reference for Airbus A320 (engine CFM56) (European Environment Agency 2016). This aircraft type was chosen as it was used in 55% of the studied flights. Less than 1% of the studied flights were performed with a large type of aircraft and the rest of the flights were represented mostly by regional class. The adapted emission factors per minute of taxiing are shown in Table
Table 1: Emissions factors per minute of taxiing.
| Type |
Factor, kg/min |
| Fuel consumption |
14.52 |
| NOx emission per min, F NO |
0.065196 |
| CO 2 emission per min, F CO |
1.7604 |
Assuming certain emission factors in this paper is made to get a general estimation of the two-module approach application impact on airport emissions. Nevertheless, it is considered to perform a more detailed calculation in the future, accounting for different emission factors for all present types of aircraft, when more actual data on aircraft specifications become available
At the time of performing this study, there was no information available about exact or historical standassignments in MEX. Therefore, the two-module approach generated assignments were compared to arandom last-minute assignment, generated directly during every simulation run. A random last-minuteassignment allocates a flight during simulation to any suitable stand available at the moment of aircraftstarting landing approach. That means that any suitable stand not occupied at the decision moment can bechosen regardless of its taxi distance to the runway. As the choice is made randomly, every simulation runresults in different usage of stands. As there is no preliminary planned assignment in such last-minuteallocation, it is considered that the effects of schedule disruptions on stand usage are minimized and thereis less possibility for assignment conflicts. Although, it is not estimated at what environmental cost theseeffects are minimized. In this section, the effects of such last-minute random allocation on the taxi-relatedemissions are estimated and compared to a proactive allocation planning, performed by the two-moduleapproach. Additionally, to trace the effects of schedule deviations on taxi-related emissions, simulationscenarios containing both on-time and disrupted arrivals were included in this study
An overview of the defined stand assignment scenarios is presented in Table 2. These scenarios can be described as follows:
- Scenario A. Base case. It represents an ideal situation with all flights arriving on time, stand assignment generated only with the use of Module II (i.e. optimized allocation without considering deviations).
- Scenario B. Stand assignment generated only with the use of Module II (i.e. optimized allocation without considering deviations). The flights arrived with arrival time deviations, generated based on arrival time deviation distributions learned in Module I.
- Scenario C. Stand assignment generated considering the expected delay with the use of both Module I and Module II. Flights arrived with arrival time deviations, enerated based on arrival time deviation distributions learned in Module I.
- Scenario D. Arriving flights are assigned to stands using last-minute random allocation. Flights arrived on time, according to the schedule
- Scenario E. Arriving flights are assigned to stands using last-minute random allocation. Flights arrived with arrival time deviations, generated based on arrival time deviation distributions learned in Module I.
Table 2: Stand assignment scenarios.
| Scenario name |
Schedule disruptions |
Schedule disruptions considered |
Assignment optimization |
Assignment generation |
| A |
– |
– |
YES |
MODULEII |
| B |
YES |
– |
YES |
MODULEII |
| C |
YES |
YES |
YES |
Two-module |
| D |
YES |
YES |
YES |
Random last- minute |
| E |
YES |
– |
– |
Random last- minute |
The objective of this paper is to discover the hidden potential for the reduction of taxi-related emissionsthrough stand assignment optimization. And as has been observed in the analysis of the generatedassignment in section 3.3, the current distribution of domestic and international areas in the terminals has aconsiderable influence on the assignment results and therefore on the level of taxi-related emissions.Therefore, the relaxation of some restrictions of MEX was considered to verify if such action can bring anybenefit to the environmental footprint of real-life stand assignment operations. Therefore, it has been decided to manipulate some of the available assignment restrictions and therefore come up with newassignment policies, that would not require major airport facilities reconstruction. The only requirementremaining strict for all simulated assignment policies is the requirement of assignment of large aircraft onlyto the specially equipped stands. The new assignment policies were compared to the original policy, whichcontains strict assignment constraints, through the series of experiments, simulating scenarios A-E undereach of the defined policies. In such a way for every assignment policy, the performance of the two-moduleapproach under on-time and disrupted arrivals were evaluated and compared to the random last-minuteallocation. The defined assignment policies include the following:
- Group I – base case experiments. Stand assignment generated according to the original set of assignment restrictions with strict adherence to the designated terminal and international/domestic zone.
- Group II – aircraft are allocated to any available stand in the originally planned terminal. This means that both international and domestic flights can be allocated to the same stand.
- Group III – aircraft may choose stands in any terminal but must obey the designated zone policy. This means that a domestic flight must be assigned to the domestic zone but can be assigned to the domestic zone of any terminal.
- Group IV – aircraft can be assigned to any zone of any terminal. This is a layout restrictions-free assignment policy that allows getting closer to the minimum possible taxi distance and taxi-related emissions for the studied flight schedule.
- Group V – Terminal 1 is fully designated for domestic flights. This means that even if a flight was originally planned to Terminal 2, in case if it is domestic it will be assigned to Terminal 1.
- Group VI – Terminal 1 is fully designated for international flights. This means that even if a flight was originally planned to Terminal 2 if it is international it will be assigned to Terminal 1.
Using the same data to learn Bayesian distributional models for schedule disruptions and to generate simulation experiments stochasticity can be considered as a limitation of this paper. Nevertheless, the main goal of the proposed approach is to mitigate the negative impact of schedule disruptions on the airport environment, not to predict the exact delay or early arrival time for the scheduled flights. By considering a certain probability interval in the assignment planning, we intend to provide a tool for influencing stand allocation robustness. With a bigger probability interval, more perturbations can be considered; however, it might reduce stand resources capacity and thus, can be seen as a limitation for some congested airports. Smaller probability intervals would result in smaller stand blocking times but might increase the number of aircraft that might wait for the stand availability. This trade-off is not discussed in this paper, although will be explored in future research
For each assignment policy, experiments A-E were executed with 30 replications each. Each replication had a duration of 7 days plus extra hours for arrival schedule deviations. The next section presents and discusses the results of the performed experiments
