www.nature.

com/scientificreports

OPEN Integrated optimization

of scheduling for unmanned
follow‑me cars on airport surface
Dezhou Yuan , Xinping Zhu *, Yajun Zou  & Qing Zhao

To promote the application of automated vehicles in large airports, in this study, we present an
integrated optimization method for scheduling Unmanned follow-me cars. The scheduling process
is divided into three phases: Dispatch, Guidance, and Recycle. For the Dispatch phase, we establish
a vehicle assignment model, to allocate the vehicle resource equitably. For the Guidance phase,
we offer an quantitative way, to measure the spacing between Unmanned follow-me car and
aircraft. To optimize the efficiency of airport operation in the three phases and ensure safety, the
collaborative planning model, and the conflict prediction model are established. An improved grey
wolf optimization algorithm is adopted to enhance the convergence speed and generalization
performance. A case study at Ezhou Huahu Airport in China demonstrates the effectiveness of the
methods. The results show that the model of collaborative planning can make the balance of path
selection, Unmanned follow-me car’s working time, and departure sequence. The convergence speed
of the improved algorithm has been increased by 18.75%. The inequity index of vehicle assignment is
only 0.015731, and the spatiotemporal distribution of conflicts is influenced by the airport’s surface
layout.

With the recovery of civil aviation after the epidemic of COVID-19, the introduction of unmanned vehicles
at airports is considered as an innovative measure to improve airport operational efficiency and reduce costs.
These vehicles, equipped with advanced sensors, controllers, enable autonomous driving functions with complex
environment perception, intelligent decision-making, and collaborative intelligence. They have become the new
generation of airport equipment for operations and ­support1. However, airport operation is a critical issue about
safety, and it requires full consideration. Now that a single unmanned vehicle, has the capability for automatic
­control2, the overall risk control of the aviation transportation system in the hermetic environment of airport
surface has become the focus. How to conduct scheduling of the fleet scientifically is a major problem for airports.
To balance the security and efficiency of the fleet, scholars have conducted research on the scheduling problem
of unmanned vehicles in the airport scene. Basically, the scheduling process involves path planning and vehicle
­assignment3. For path planning, an optimized combination model is introduced. Because this model needs to
discuss the situation in different velocities of the vehicles, it was solved with the heuristic, such as hill climbing
particle swarm optimization (HC-PSO) ­algorithm4. For the other, the utilization ratio is considered as a kind
of index, to evaluate the equity of vehicle assignment. A dynamic programming model with various charging
strategies meets the needs of allocation for vehicle r­ esource5. Generally, the scheduling problem is a classic
vehicle routing problem with time windows (VRPTW). Different models and algorithms have been explored
to address this challenge, including linear programming (LP), combinatorial optimization, graph theory, and
network ­analysis6. Common algorithms include exhaustive methods, C-W savings algorithms, column genera-
tion algorithms, heuristic search, genetic algorithms (GA), and particle swarm optimization (PSO) algorithms.
However, research on unmanned vehicle scheduling on airport surface mostly focused on electric tractors, which
are subject to traction speed l­ imitations7 and do not meet the requirements of large-scale airports. There is a lack
of research on the scheduling process of unmanned follow-me cars (UFMCs).
Regarding VRPTW on airport surface, integer linear programming (ILP) algorithms is utilized to determine
the optimal allocation of ground support equipment (GSE) based on cost and time c­ onstraints8. The sorting of
flight support operations for airport service vehicles is improved by GA, to optimize the structure and quantity
of service vehicles. The objective is to enhance the efficiency of service vehicle u­ tilization9. An energy consump-
tion model for GSE is developed by the adaptive large neighborhood search (ALNS) algorithm, and provides
scheduling solutions for scenarios involving both fuel-powered and electric vehicles, addressing the challenge of

College of Air Traffic Management, Civil Aviation Flight University of China, Deyang 618307, China. *email: zxp@
cafuc.edu.cn

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 1

Vol.:(0123456789)
 www.nature.com/scientificreports/

minimizing energy c­ onsumption10. Dispatching issue for electric GSE in the apron is addressed by hybrid neural
networks (HNN), contributing to the overall improvement in e­ fficiency11. An intelligent dispatching model for
support vehicles is solved by GA, with a focus on optimizing the allocation of GSE and e­ fficiency12. A theoretical
system based on ILP is developed to solve the assignment problem of airport operational units, providing an
effective framework for optimizing the allocation of operational units within the ­airport13. Collaborative decision-
making for parking allocation and shuttle scheduling is successfully implemented through column generation
(CG) algorithm, aiming to improve the efficiency of parking allocation and shuttle scheduling p ­ rocesses14. An
objective function that minimizes operating costs and flight delay costs and optimizes GSE scheduling is devised
with the help of ILP, to reduce operational costs and minimize flight delays through GSE ­scheduling15. Heuristic
search methods are employed in a real-time scheduling decision system to efficiently assign ­GSE16.
Regarding aircraft path planning on airport surfaces, an ILP mathematical model for conflict-free aircraft
trajectories in various traffic density scenarios is established with a rolling time domain calculation s­ trategy17.
A mixed ILP (MILP) model for gate allocation is developed by combining operation modes of multiple runway
airports, aiming to optimize gate a­ llocation18. Heuristic methods are implemented to search for the optimal speed
profile for ground taxiing process, providing valuable references for aerodrome control ­units19. Optimal taxiing
speed profiles were generated with an exhaustive method based on required time of arrival (RTA)20. Surface
routing planning problems are also studied through multi-agent s­ imulation21.
In conclusion, there are various optimization techniques employed in scheduling UFMCs on airport surfaces,
including ILP, GA, ALNS, HNN, and heuristic search methods. Additionally, mathematical models such as ILP
and MILP are utilized for path planning on airport surface. Some professional concepts such as RTA, energy
management and flow of transportation management are introduced, that makes mathematical models meet the
demands of business processes. However, further exploration is needed to develop more efficient and accurate
optimization techniques for the scheduling of UFMCs on airport surface.
While previous research has provided insights into the issue of scheduling on airport surfaces, these studies
have some limitations. Firstly, they mainly consider vehicles for small to medium-scale operations, typically
involving up to 50 flights, which may not meet the demands of larger-scale vehicle scheduling. Secondly, although
the vehicle assignment models may generate satisfactory solutions, they often fail to take uncertainties and flight
support requirements into account. Finally, the evaluation of potential movement conflicts is not thoroughly
addressed in these studies. The conflicts are either treated as disturbances or some mitigation suggestions are
provided.
On the other hand, in advanced surface movement guidance and control systems (A-SMGCS) Level 4 and
above, the surface movement of aircraft relies on three functional modules: target surveillance, conflict alert, and
routing. However, the conventional method of surface taxiing guidance with lights of centerline on the taxiway
proves to be expensive in terms of maintenance. It is not suitable for large and busy airports. To address this issue,
an alternative approach within a generalized A-SMGCS implementation framework is proposed, which utilizes
UFMCs as a replacement for taxiing guidance. Figure 1 shows the basic working principle of this approach, where
the UFMC scheduling system integrates with the existing functional modules. In this concept, the aircraft follows
the UFMC during taxiing, requiring close attention from the pilot for guidance. The advantage of UFMCs over
traditional manual follow-me cars lies in their ability to formulate efficient vehicle scheduling schemes, enabling
prompt departures from the parking lot and seamless integration with guidance tasks. Furthermore, UFMCs
overcome the drawback of slower towing speeds associated with electric tractors, ensuring the efficiency of

Figure 1.  Working principle of the UFMC scheduling system.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 2

Vol:.(1234567890)
www.nature.com/scientificreports/

airport surface operations. Therefore, this paper aims to study the integrated optimization of UFMC scheduling
to address this guidance requirement.

Problem description
The integrated scheduling process for UFMCs to facilitate aircraft taxiing guidance is illustrated in Fig. 2. This
process comprises vehicle trajectory planning and vehicle assignment. Vehicle trajectory planning involves
generating optimal trajectories for the UFMCs across three phases: dispatch from the parking lot to the start of
guidance (Dispatch), implementation of the guidance process (Guidance), and return to the parking lot after
completing the guidance (Recycle). These trajectories are generated based on specific safety and efficiency objec-
tives. Each vehicle’s trajectory consists of a sets of position coordinates x, y along the route, corresponding

velocities (v), and timestamps (t). Vehicle assignment refers to the selection of a specific UFMC from a pool of
candidates based on certain optimization objectives to perform ground taxiing guidance for the corresponding
flight.
As can be observed, the precise planning of UFMCs in terms of starting, stopping, and speed profiles enables a
higher level of refinement compared to manually operated follow-me cars. This allows vehicles to promptly reach
the starting point of the guidance task and proactively plan to avoid potential conflicts during their operations.
For the convenience of discussion, the UFMC and the aircraft being guided by it are considered as a “Guidance
Unit (GU)” in the surface movement scenario.
From Fig. 3, during the Guidance phase, there exists a following relationship between the aircraft and the
UFMC. The spacing between them affects the spatial occupancy of the GU. It also influences the scheduling
decisions. To address this, a definition and a quantification method for the distance are provided:

Definition 1 (Guidance following spacing) Within the GU, the spacing L from the front of the aircraft, to the
rear of the UFMC.

As shown in Fig. 4a, the measurement of L is similar to the concept of Following Spacing in transportation
­engineering22:
lA lV
L = dAV − − (1)
2 2

Figure 2.  The integrated scheduling process for UFMCs.

Figure 3.  Schematic diagram of a GU.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 3

Vol.:(0123456789)
 www.nature.com/scientificreports/

where lA and lV are the lengths of the aircraft and the UFMC, respectively. dAV represents the distance between
their geometric centers.
During the following process, the visual influence to the pilot must be considered in order to comprehensively
evaluate the safe guidance following spacing for each type of aircraft. As depicted in Fig. 4b, there exists the
following relationship between the visual spacing dvisual in the cockpit, and the guidance following spacing L
is calculated by the Pythagorean theorem.

Model formulation
The scheduling model for UFMCs consists of two modules: vehicle trajectory planning and vehicle assignment.
Figure 5 shows the details about this model:

• The vehicle trajectory planning module comprises the method for guidance trajectory deduction and a
conflict prediction model. The former is responsible for rapidly deducting and generating a solution set of
trajectories based on pre-planned vehicle routes and delivery times, with the input of guidance tasks. The
latter is used to evaluate potential conflicts and determine the feasibility of the solution set. Finally, all feasible
solution sets are integrated to generate an optimal solution set of trajectories.
• The vehicle assignment module transforms the optimal into corresponding guidance tasks, employing the
UFMC assignment model to achieve the assignment function. Ultimately, the optimal scheduling scheme,
i.e., the timetable of each guidance task, is obtained.

Collaborative planning model for surfaceguidance trajectories

The set of flights to be guided denoted as G = g1 , g2 , . . . , gs , where s means the total number 
of guidance

tasks, i.e., the total number of flights. The three phases of the guidance task are denoted as P = p1 , p2 , p3 ,

corresponding to the Dispatch phase, Guidance phase, and Recycle phase.

Objective function
To meet safety and efficiency requirements, the objective function is designed from the following perspectives:

• Minimizing the number of conflicts

The number of conflicts be denoted as Nconflict , and set the objective to minimize the number of conflicts:
min NConflict (2)

Figure 4.  Measurement method for guidance following spacing. (a) Vertical view. (b) Lateral view.

Figure 5.  Framework of the scheduling model for UFMCs.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 4

Vol:.(1234567890)
www.nature.com/scientificreports/

• Minimizing the working time of UFMCs

Minimizing the working time of UFMCs means reasonable route planning, and reducing the ineffective waiting
time of UFMCs at the gap between two phases as possible. Therefore, the objective is to minimize the working
time of UFMCs:
s 
 3
min tij (3)
i=1 j=1

Here, i represents the i-th guidance task in the timetable, j represents the phase at which the UFMC is located,
and s is the total number of flights in the timetable. The working time of UFMCs is the sum of the working times
in each phase of the guidance tasks, and it is measured in seconds. Therefore, it could reach the magnitude of
100,000. Meanwhile, the number of conflicts is up to 1000 extremely.
To achieve the comprehensive optimization of the two objectives, the range of working time of UFMCs is
normalized to [0,1] as follows:

s 3
T = � �t

ij
i=1 j=1 (4)
 T′ = 1 − 1

T

Here, the working time of UFMCs is transformed into T ′ , to remove the influence of dimension while ensuring
the monotonicity of the indicator.
Therefore, the objective function for vehicle trajectory planning is:
min Z = θ1 T ′ + θ2 Nconflict (5)
While setting θ1 = 1, the integral part of the objective function value Z maps the number of conflicts on the field,
and the decimal part maps the total working time of UFMCs on airport surface. The value of θ2 is determined
by the scale of the airport surface, ensuring that ∀Nconflict ∈ N , θ2 Nconflict is one order of magnitude larger than
θ1 T ′23, achieving simultaneous optimization of the two components.

Constraints

• Constraints of routing for UFMCs

For the three phases of vehicle guidance, the UFMC scheduling system has prestored one shortest route and
(k∗ − 1) sub-optimal routes for each pair of origins and destinations. During the process of scheduling, the system
selects one from the k∗ candidates as the route of UFMC, i.e.,
Rij  k∗ , ∀i ∈ G, j ∈ P (6)
where Rij is an integer decision variable that represents the number of the route. The subscript i and j indicate
that route selection is performed for each guidance task and each phase in the trajectory planning module.

• Constraints of time windows for UFMCs

The aerodrome control unit and the Airport Operation Center (AOC) adjust the delivery time windows of the
UFMCs through collaborative decision-making throughout the flight process to achieve integrated optimization
of scheduling of UFMCs and control of flights:
γij  δj , ∀i ∈ G, j ∈ P (7)
where γij is the decision variable, corresponding to the delivery time adjustment for the i-th guidance task in
phase j. δj represents the upper limit of adjustment for each phase.
Taking guiding flight i as an example, the delivery times for the three phases of the UFMC are shown in
Table 1. ETA and ETD are obtained from the flight schedule. They represent the Estimated Time of Arrival and
Estimated Time of Departure, respectively. LDR is the time from the arrival aircraft touching down, to reaching
the starting point of the Guidance. LUP is the time from the departure aircraft waiting at the runway entrance
to having clearance of lining up the runway, respectively. CRS represents the time taken for the UFMC to pass
through the taxiway on the apron, while UT is the time for the UFMC to make a U-turn in front of an arrival
aircraft.

• Constraints of domain

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 5

Vol.:(0123456789)
 www.nature.com/scientificreports/

Flight Type Phase Regular delivery time (RDT) Actual delivery time (ADT)
Arrival 1 ETA+LDR-ti1-UT RDT-γi1
Arrival 2 ETA+LDR RDT-γi2
Arrival 3 ETA+LDR+ti2+CRS RDT-γi2 + γi3
Departure 1 ETD-LUP-ti2-ti1-CRS RDT-γi1 − γi2 − γi3
Departure 2 ETD-LUP-ti2 RDT-γi2 − γi3
Departure 3 ETD-LUP RDT-γi3

Table 1.  Delivery time of UFMC guiding flight i in three phases.

To ensure the meaningfulness of each variable, it is necessary to define their domain, including the number of
conflicts as a natural number, the route number of the UFMC as a positive integer, and the non-negativity of the
UFMC working time and delivery time adjustment:
NConflict ∈ N (8)

k, Rij ∈ N+ , ∀i ∈ G, j ∈ P (9)

tij , γij , δj  0, ∀i ∈ G, j ∈ P (10)

Trajectory deduction
The purpose of guidance trajectory deduction is to deduce the trajectory of an UFMC or GU, for each guidance
task at different phases based on a predetermined route and the delivery time. For UFMCs, precise control to
parameters can be achieved based on the setting of guidance speeds in different zones, vehicle performance
parameters, as well as road conditions, enabling accurate trajectory deduction to support the prediction of
potential conflict in trajectories.

Definition 2 Trajectory of UFMC The coordinates the UFMC’s geometric center passes through, along with
corresponding timestamps and velocities.

Definition 3 Trajectory of GU The coordinates the GU’s geometric center passes through, along with correspond-
ing timestamps and velocities.

Since the preplanning process has generated several candidate routes, in order to accurately match the UFMC
with the guidance requirements, further algorithm design is required, to generate high-precision velocity profiles
and achieve synchronization between the actual delivery time and the start of the timestamp.
The input to the algorithm of Speed Profile Generation for UFMC is a set of nodes formed by a single candi-
date route. Then, correct the speed limit and determine the safety throttle/braking distance for each road segment,
by basic kinematic equations. Finally, generate the final velocity profile:

vn = vn−1
n
sn−1 = vn−1 (tn − tn−1 ) (11)

′

vn = vn
n −s
sn−1 safe vn −vn−1 (12)
tn − tn−1 = vn + a

2
vn2 − vn−1 n

= 2asn−1
n vn +vn−1 (13)
sn−1 = 2 (tn − tn−1 )

The algorithm is showed as following:

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 6

Vol:.(1234567890)
www.nature.com/scientificreports/

Algorithm 1.  Speed Profile Generation for Unmanned Follow-me Car.

The process for the guidance trajectory deduction is as follows:

• Input the current phase j,the number

  of parking stand where flight i is located, and the selected route Rij .
Generate the set of nodes x0 , y0 , x1 , y1 , x2 , y2 , . . . , xt , yt corresponding to the guidance route;
   
• Input the delivery time adjustment γij and determine the delivery time, denoted as t0, according to Table 1;
• Traverse the nodes on the trajectory, and determine the turning points via the dot product formula:
� · BC
AB �
cos α =    (14)
� BC
AB � 

Where point B is the point to be determined, and points A and C are the previous and next nodes, respec-
tively. If α > 60◦, point B is identified as aturning point.
• Determine the speed limits v1 , v2 , . . . , vt for each point based on actual operational conditions:
′ ′ ′

Figure 6.  Smoothing for segments at turning areas.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 7

Vol.:(0123456789)
 www.nature.com/scientificreports/

vi′ = vnode (node ∈ EApron , ETaxiway , ETurn ), ∀i ∈ [1, t] (15)

Where vnode represents the speed limit at the next node, i.e., the current section’s speed limit. EApron, ETaxiway ,
and ETurn represent the apron (including service roads), the taxiways in the maneuvering area, and the turn-
ing points, respectively;
• Call Algorithm 1 to generate the velocity profile {(v1 , t1 ), (v2 , t2 ), . . . , (vt , tt )}:
• Smooth the segments at the turning areas as shown in Fig. 6. ∀node ∈ ETurn, adjust the timing as follows:
�α�
 �x = �r tan

α
� 2
�s̃ = � 2 �r (16)
tTurn ′ = tTurn − 2(�x−�s̃)

vTurn

Where α is the turning angle, r is the turning radius, x is the advance distance for the turn, s̃ is half of the
arc length corresponding to the turning
  angle,
 and tTurn is the adjusted timing
′
 for passing the turning point.
• Output the guidance trajectory:T i, j = xn , yn , vn , tn |n = 0, 1, 2, . . . , t .

The guidance trajectory T i, j provides information about the working time of UFMC:
 

tij = tt − t0 , ∀i ∈ G, j ∈ P (17)
Moreover, the guidance trajectory T reflects the situation of the UFMC on surface, and serves as the foundation
for the conflict prediction model.

Conflict prediction
A conflict prediction model based on protected zones is established by using the coordinates of each position
in the trajectory as the geometric centers. The properties of a protected zone are determined by trajectory’s
parameters of corresponding position, enabling the prediction of conflicts. Among commonly used protected
zones, the elliptical demonstrates good prediction accuracy and intuitiveness.
The lengths of the semi-major axis “a” and the semi-minor axis “b” of the elliptical protected zone are cal-
culated as following:
 � �
 a = ltarget + v2 · ksec urity
2 2a
� � d (18)
 b = wtarget · ksec urity
2

Here, ksec urity represents the safety factor. ltarget and wtarget denote the length and width of the UFMC or GU,
respectively. During the Guidance phase, the moving target refers to the GU. During the Dispatch or Recycle
phase, the target refers to the UFMC:

l + lA + L if target is a Guidance Unit
ltarget = V (19)
lV else


wA if target is a Guidance Unit
wtarget =
wV else (20)

Here, wA represents the wingspan of the guided aircraft, and wV represents the width of the UFMC.
The purpose of setting the safety factor ksec urity is to minimize the possibilities of both false alarms and
missed alerts in conflict prediction. Figure 7 shows the scenarios of different safety factors corresponding to the
protected area.
The black protected zone represents a safety factor of 1, which cannot envelop the entire moving target most of
the time, potentially leading to missed alerts. The red protected zone corresponds to a safety factor of 1.2, which

Figure 7.  Schematic diagram of a protected zone.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 8

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 8.  Schematic diagram of a typical situation of conflict.

can envelop the entire moving target and minimize the probability of false alarms when potential conflicts are
predicted. The ­Reference24 has already discussed incidents of wingtip collisions on the apron, it is sufficient for
their model to recognize and deal with such incidents. Therefore, our model eliminates the wingtip clearance ε
proposed by Giuseppe S [7], and uses the length of the GU instead of the length of the aircraft in their tractor
problem to ensure that there is sufficient space for our GU within the protected zone.
Figure 8 visually shows a typical scenario of conflict between GUs. The elliptical protected zone could accu-
rately represent the spatial occupation of GU, and maintain safe spacing both longitudinally and laterally. To
reproduce the protected zone, and determine whether the ellipses overlap or not, We have formulated the ellipti-
cal equation as following:
   2    2
(x − xn ) sin θ − y − yn cos θ (x − xn ) cos θ + y − yn sin θ
+ =1 (21)
a b

Subsequently, the Algorithm 2 is used to output the number of conflicts, denoted as NConflict , thereby completing
the conflict prediction.

Algorithm 2.  Conflict Prediction Algorithm.

Model for UFMC assignment

Compared to traditional manually-operated vehicles that rely on human experience for scheduling, operation,
and maintenance, UFMCs overcome the limitations of manual driving through fully automated control. Build-
ing upon precise trajectory planning, further considerations for vehicle maintenance and optimal allocation of
vehicle resources are necessary. In this regard, an UFMC assignment model shown in Fig. 9 is designed in the
UFMC scheduling system, wherein the optimal solution set of trajectories is transformed into specific guidance
tasks, to achieved the function of assignment through ILP.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 9

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 9.  Framework of the model of UFMC Assignment.

Considering that the source assignment problem can attribute to 0-1 programming. According to the general
method in management of civil a­ viation13, we designed a model for vehicle assignment as following:
s 
k + 1 2  kl=1 (k − l + 1)xil
min G = − 3  
k k j=1 tij + η
i=1

s.t.
k
�
xil = 1, ∀i ∈ G
l=1
xol + xpl � 1, ∀l ∈ F
�
ao � ap + η + 3j=1 tpj
�
�3
ao + η + j=1 toj � ap
 q s dil − �i∗
 � ��� � �
3
i=1 i=1 j=1 tij + η xil � 0
, ∀i∗ ∈ G, l ∈ F
 q �s dil − �i∗
��� � �
3
i=1 i=1 j=1 tij + η xil � q + qreserve

xol + xrl � 2 − dol , ∀l ∈ F

�
ao � ar + η + C + 3j=1 trj
�
� �
�3 , ∀l ∈ ∁F fk
ao + η + C + j=1 toj � ar
�
1 if vehicle l guides flight i
xil =
0 else
�
1 if vehicle l is charged after flight i
dil =
0 else
xil − dil � 0, ∀i ∈ G, l ∈ F

Model assumptions

• The UFMCs at the end of the Recycle phase must return to the parking lot where they started, in order to
perform the next guidance task.
• A buffer time of η is allocated between two guidance tasks to ensure orderly traffic flow in and out of the
parking lot, or transitioning to the corresponding parking lot for the next guidance task.
• A vehicle work continuously until its battery level reaches a low state, then return to the parking lot for a
full recharge. The charging time C and the safe endurance q are fixed values specific to each type of vehicle.
Factors such as battery decay and seasonal variations are ignored, but the buffer time η is included in the
continuous working time.

Let F = f1 , f2 , . . . , fk represent the set of k UFMCs in the airport.

 

Objective function
This model aims to achieve the most equitable vehicle assignment, so we adopt the Gini coefficient as the evalu-
ation metric. In the field of economics, a Gini coefficient between 0.3 and 0.4 is considered ­fair25. However,

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 10

Vol:.(1234567890)
www.nature.com/scientificreports/

considering that the working time of UFMCs is a “negative asset”, the Gini coefficient is referred to as the
“assignment inequality index” in this context. Therefore, the objective function is set to minimize the assignment
inequality index of the working time for UFMCs:
k
 k

min G = Pl Yl + 2 Pl (1 − Vl ) − 1
l=1 l=1
s k (22)
k+1 2  − l + 1)xil
l=1 (k
= − 3  
k k j=1 tij + η
i=1

Where Pl = and Yl represents the proportion

k,
1
of working time for the l-th vehicle to the total working time,
and Vl is the cumulative distribution of Yl . 3j=1 tij denotes the accumulated time for the three phases of a single

guidance task. xil is a binary decision variable indicating whether vehicle l guides flight i.

Constraints

• Constraints of flight coverage

Each flight should be assigned to one UFMC ONLY:

k

xil = 1, ∀i ∈ G (23)
l=1

• Constraints of consecutive guidance tasks

The time windows of guidance tasks assigned to the same vehicle should not overlap. Let o and p represent a pair
of guidance tasks corresponding to flights where their time windows overlap each others, then the constraints
can be formulated as follows:
xol + xpl  1, ∀l ∈ F (24)
The condition that the time windows of guidance tasks o and p overlap can be expressed as follows:

ao  ap + η + 3j=1 tpj

3 (25)
ao + η + j=1 toj  ap

Here, ao and ap correspond to the actual delivery times (ADT) of guidance tasks for flights o and p, respectively.
Usually, guidance tasks are sorted by ADT. Fig. 10a shows the overlap situation of time windows between
adjacent index of guidance tasks.

• Constraints of vehicle electricity

Figure 10.  Overlap contribution. (a) For time windows of working. (b) For time windows of charging.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 11

Vol.:(0123456789)
 www.nature.com/scientificreports/

Constraints of vehicle electricity ensure that the quantity of electricity of the UFMCs meets both upper and lower
limit requirements during operation and forms a closed loop for charging:

 q s dil − �i∗
 � ��� � �
3
i=1 i=1 j=1 tij + η xil  0
 q �s dil − �i∗
��� � �
i=1 i=1
3
j=1 tij + η xil  q + qreserve
(26)
, ∀i∗ ∈ G, l ∈ F
Here, dil is the binary representing whether vehicle l requires charging after completing task i. qreserve is a backup
of the quantity of electricity that ensures the UFMC can safely return to the parking lot after operating for a
continuous duration of q.

• Constraints of charging

For single vehicle, the guidance task should not have conflict with the charging process:
xol + xrl  2 − dol , ∀l ∈ F (27)
The conflict condition between the time windows of guidance tasks o and r can be expressed as follows:

ao  ar + η + C + 3j=1 trj

(28)
ao + η + C + 3j=1 toj  ar


Figure 10b shows the overlap situation between the time windows of a guidance task with charging demand,
and its follow-up tasks.

• Constraints of vehicle sorting

As Fig. 11 is shown, the calculation of assignment inequity index is related to the equity curve and the Lorenz
curve. Both of them are formulated on the basis of cumulative distribution of UFMC working time. Therefore,
each element of the set of vehicles F = f1 , f2 , . . . fn should be sorted in ascending order based on its working

time:
s 
 3 s 
 3
   
tij + η xil � tij + η xi(l+1)
i=1 j=1 i=1 j=1 (29)
 
, ∀l ∈ ∁F fk

• Constraints of domain

1 if vehicle l guides flight i
xil =
0 else (30)

Figure 11.  Calculation of assignment inequity index with equity curve and Lorenz curve.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 12

Vol:.(1234567890)
www.nature.com/scientificreports/


1 if vehicle lis charged after flight i
dil =
0 else (31)

xil − dil  0, ∀i ∈ G, l ∈ F (32)

Solution algorithm
In this section, we design corresponding solution algorithm based on characteristics of models mentioned, and
construct an overall solution framework for the scheduling problem for UFMCs.
As Fig. 12 is shown, the solution process for the scheduling model for UFMCs consists of three steps: prepara-
tion of basic data, generation of trajectory on surface, and vehicle resource allocation.

• In the step of basic data preparation, a comprehensive graph model G = (V , E) is generated, where the edges
E correspond to various road sections in the scene, including EApron (the service roads, taxiways and taxi links
within the apron), ETaxiway , taxiways in the maneuvering area, as well as ETurn, the various intersections and
turns. The vertices V represent the endpoints of the edges, including key points like turning points, entrances/
exits, and crossings. Then, generate a set of candidate routes for each pair of origins and destinations (ODs)
on airport s­ urface26.
• In the step of generation of trajectory on surface, the decision variables Rij , γij i ∈ G, j ∈ P are trans-
  
formed into the assigned routes and delivery times in three phases of each guidance task. A function of fitness
is formed by integrating the collaborative planning model for surface guidance trajectories and the conflict
prediction model based on protection zones. The optimal set of surface guidance trajectories are formed
iteratively by the IGWO algorithm.
• In the step of vehicle resource allocation, the optimal set serves as the input for the UFMC assignment model.
The model is solved directly using a solver to output the allocation scheme for vehicle resource.

The complexity analysis shows that the scale of feasible solution of the collaborative planning model for surface
guidance trajectories is extremely large, with 2s possible combinations of ONLY the route components. When
the number of daily flights exceeds 50, there is likely to be a combinatorial explosion problem. The scale of the
decision variable for the UFMC assignment problem is (2ns), which is a relatively manageable classic ILP problem
and does not require a heuristic algorithm additionally.
Therefore, to address the collaborative planning model for surface guidance trajectories, an IGWO algorithm
based on integer encoding is designed to improve the efficiency of generation for the optimal set of surface
trajectories.

Figure 12.  Framework of solution method.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 13

Vol.:(0123456789)
 www.nature.com/scientificreports/

Encoding strategy and initialization

When we apply this algorithm, it is necessary to define the meaning of each dimension in the position vector of
a grey wolf individual. Based on the decision variables of the collaborative planning model, targeting guidance
tasks for s flights, we construct an integer space with 6s dimensions as shown in Table 2.
Considering that the UFMC scheduling system may avoid conflicts by adjusting routes, shifting time win-
dows, or using a combination of both, conflict-free solution sets would be dispersed at multiple locations, we
randomize the wolf pack initially, then integer by rounding to construct the integer space. This serves as the
start for iteration and optimization.

Fitness function
The calculation of the fitness function needs to consider the total working time on airport surface, which is based
on the selected route, and needs to deduct the situation of potential conflict between the selected route and the
variation in delivery time. Therefore, the fitness function f is defined as follows:
 
1
f =θ1 1 − s 3
i=1 j=1 tij
86400
(33)
   
+ θ2 card node ∈ Ellipsest ∗
t ∗ =1

In this equation, the first term, which relates to the target parking number and selected route, can be obtained
by from the pre-computed values of tij based on the given timetable of flights and information of grey wolves.
The second term requires traversing each unit of time for movement on the surface, and ∀node,we need to count
the number of times a node falls in different ellipses.

Enhancing global search

In the local search, the slight decimal-level changes in the positions of the wolf pack in the traditional GWO
algorithm are not suitable for integer encoding, so we adopt the IGWO algorithm proposed by Yang ­Z27 to
enhance the prominence and efficiency of global search during the iteration process. The formula for updating
the convergence factor is given as follows:

(34)
t
a′ = 2e− T
Where the improved convergence factor a′ starts
from the default value of 2 and exponentially decays to 2e
with
the iteration process. When approximately 69.3 percents of the iteration process is completed, the improved
algorithm enters the local search, and the global-to-local search ratio shifts from 1:1 to 7:3.

Improvement for local search

For the local search, we adopt a probabilistic method for updating ­position28:

 X 1 rc ∈ �0, 31 �
 � �

X(t + 1) = X 2 rc ∈ � 31 , 23� (35)

X 3 rc ∈ 32 , 1


Here, rc is a random number that satisfies U(0, 1). The formula for updating position based on integer space not
only reflects the dynamic process of the wolf pack following the leader, but also ensures the meaningfulness of
the operation of updating position for X(t + 1) in the local search.

g1 g2 ... gs Explanation
X i1 1 1 1 Route NO. at Phase 1
X i2 2 1 1 Route NO. at Phase 2
X i3 1 2 1 Route NO. at Phase 3
X i4 55 6 45 Time adjustment at Phase 1
X i5 488 20 271 Time adjustment at Phase 2
X i6 33 276 54 Time adjustment at Phase 3

Table 2.  The vector of grey golf X with integer encoding.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 14

Vol:.(1234567890)
www.nature.com/scientificreports/

IGWO algorithm

In summary, the flow of the IGWO algorithm is shown as following:

Algorithm 3.  IGWO Prediction.

Preparation of data
Ezhou Huahu Airport is a specific cargo airport in China, it is also an important experiment field for unmanned
ground support vehicles, but there is a lack of overall scheduling for large-scale fleet.Taking Ezhou Huahu Airport
for simulation experiment, we investigate the optimization of scheduling for UFMCs during a typical working
day in 2030, with 270 flights per day, and assume that runway-to-stand assignments are based on the principle
of proximity.
Before starting the experiment, we has built a simulation platform. It is divided into air traffic control (ATC)
terminal, pilot terminal, and data processing terminal. The interaction between terminals, and the details about
hardware and software are shown in Fig. 13.

Figure 13.  Schematic diagram of the UFMC’s routes.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 15

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 14.  Schematic diagram of the UFMC’s routes.

Figure 15.  Schematic diagram of optional route set for Ezhou Airport. (a) Route set for Departure. (b) Route
set for Arrival.

Basic routes
Figure 14 shows the schematic diagram of the UFMC’s routes. During phases 1 and 3 (operations on service
roads), the strategy of separating inner and outer loops is employed. In phase 2 (on-taxiway) operations, the
­ rinciple29 is followed, where alternative routes are assigned to vehicles, to
“fixed, one-way, directional, cyclic” p
prevent and alleviate congestion, and to minimize the occurrences of conflict initially. The number of alterna-
tive routes, denoted as k, is set to 2. The blue and green arrows in the figure represent Route 1 and Route 2,
respectively.
Based on the strategy for route allocation, Ezhou Airport’s UFMCs have a set of selectable routes shown
in Fig. 15. Figure 15a and b show the departure and arrival scenarios, respectively. In these figures, Route 1 is
highlighted in blue (shared with Route 2 in green) during the Guidance phase, while Route 2 is labeled in green.
The Dispatch and Recycle phases are marked by the red and cyan lines for Route 1 and 2, respectively.

Parameter Explanation Value Parameter Explanation Value

q(h) Safety endurance 12 aa [m/s2 ] Acceleration 1
qreserve(min) Reserve endurance 15 ad [m/s2 ] Deceleration 1
vTaxiway(m/s) Velocity on main taxiways 10 lcar(m),wcar(m) Length and width 4.8, 1.8
vApron(m/s) Velocity in apron 8 C(min) Required Time for Charging 45
vturn(m/s) Velocity at turns 5 L(m) Average Following Space 60

Table 3.  Parameter settings for the fleet of UFMCs.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 16

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 16.  The velocity profiles along typical routes.

Parameter settings
According to the actual operation of Ezhou Airport and the requirements of Chinese regulations [20], the
parameters of the UFMC are shown in Table 3. The UFMC is allowed to arrive in advance and wait for a maxi-
mum of 60s. The time windows for air traffic controllers to adjust the ETD and ETA of flights are 600 s and 300
s, respectively. Therefore, in the case of arrival, the upper limits of the change in delivery times for UFMCs (γi1
to γi3) are 60s, 300s, and 60s, respectively; for departure, those are 60s, 60s, and 600s, respectively.

Profiles generation
To illustrate this process, we selected Route 1, along with guarded flights at Stand 111 and 301 as examples, and
generated velocity profiles by Algorithm 1. Figure 16 shows the typical velocity profiles.
In Fig. 16, the velocity of the GU remains stable at 5 m/s, 8 m/s, and 10 m/s, corresponding to turns, aprons,
and straight sections of taxiways in maneuvering areas, respectively. As shown in the second half of each profile,
the UFMC adjusts its cruising speed, throttle, and brake, based on a comprehensive modification of the length
of each segment and the initial settings, through the application of the Algorithm 1. The velocity is changed
when the GU passes through different areas and segments, suggests that the efficiency of the UFMCs scheduling
is related to the number of turns, taxiing distance, and timing of crossing the apron area, which confirms the
importance of route selection in the trajectory planning module.

Conflicts analysis
An analysis of conflict characteristics before the optimization of scheduling was conducted. Without consider-
ing the selection of surface guidance trajectories and adjustments to delivery times, Algorithm 2 predicted 824
instances of conflicts among the trajectories. Figure 17a, b, and c show the characteristics distribution of conflicts
spatially, for the west main taxiway, the entire surface, and the east main taxiway, respectively. The characteristic
of “high in the north, low in the south” of the distribution of conflicts indicates that, when operating towards the
north, the conflicts during arrival are higher than those during departure. On the other hand, due to the coexist-
ence of passenger and cargo parking stands on the west side of Ezhou Airport, the distribution also exhibits the
characteristic of “high in the west, low in the east”.

Results and discussion

Regards to vehicle trajectory planning
After the computation of the vehicle trajectory planning module, the UFMC scheduling system generated a set
of conflict-free guidance trajectories, with a total working time of 170,463s. The surface guidance trajectory is
mainly planned collaboratively in three aspects: route selection, UFMC’s working time, and delivery sequence,
achieving the effect of optimization. Among them, the distribution of route selection is shown in Fig. 18. The
histogram indicates that Route 1 with shorter travel time remains the top choice for UFMCs. In phases 1 to 3,

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 17

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 17.  Distribution of conflicts before optimization.The southern end of the taxiway corresponds to the
right-hand side of the horizontal axis. (a) Conflicts on the west main taxiway. (b) Conflicts on the east main
taxiway. (c) Conflicts on airport surface.

Figure 18.  Histogram for frequency distribution of route selection.

24.81 %, 21.48 %, and 22.96 % of guidance tasks respectively choose Route 2 to avoid potential conflicts at the
cost of detour.
Adjustments in UFMC working time are shown in Fig. 19. Figure 19a shows that by extending the working
time of UFMCs by 2.85%, the vehicle trajectory planning module achieves the resolution for all the conflicts.
Figure 19b and c show that sacrificing operational efficiency is targeted. For arrival guidance tasks with higher
conflict frequency, 7.15% of efficiency is sacrificed, while for departure, the change is minimal, with only a 0.66%
extension of UFMC working time.
Regarding the result of delivery sequence shown in Table 4, a significant number of guide tasks have been
adjusted by fine-tuning the delivery time to avoid conflicts. This has altered the regular delivery sequence,

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 18

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 19.  Box-plot of UFMC working time. (a) Overall distribution of working time. (b) Distribution of
working time for arrival. (c) Distribution of working time for departure.

Regular delivery sequence Actual delivery sequence Actual delivery time

1 2 0:07:57
2 1 0:07:36
3 3 0:15:14
··· ···
39 40 2:06:41
40 39 2:05:48
41 42 2:24:17
42 41 2:14:52
43 45 2:35:36
44 43 2:34:02
45 44 2:35:10
46 47 2:44:54
··· ··· ···
270 270 23:25:06

Table 4.  Result of delivery sequence. Significant values are in [bold].

Figure 20.  Convergence analysis of five algorithms. IGWO-1 ONLY incorporates the improvement defined by
Eq. (34), while IGWO-2 employs Eq. (34) and Eq. (35) uniformly in both the global search and local search. (a)
Fitness values. (b) Conflict counts. (c) Total working time of UFMCs.

ensuring safety without the need for detours. Out of a total of 270 guide tasks, 64.44% have undergone adjust-
ments in the delivery sequence.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 19

Vol.:(0123456789)
 www.nature.com/scientificreports/

Required generation finding a Total working time of

Sorties of takeoff and landing Algorithm Rate of conflict resolution[%] feasible solution set[s] UFMCs[s] Actual number of iterations
PSO 100 80 183,192 80
GWO 100 110 167,471 110
IGWO 100 65 170,463 65
270 IGWO-1 91.67 > 150 168,625 > 150
IGWO-2 80 > 150 166,155 > 150
WOA 100 28 176,057 35
NNA 81.82 > 150 178,684 > 150
PSO 100 Feasible Initially 6,843.1 2
IGWO 100 Feasible Initially 6,744.9 18
IGWO-1 100 Feasible Initially 6,801.2 10
12
IGWO-2 100 Feasible Initially 6,788.9 32
WOA 100 Feasible Initially 6,851.2 17
NNA 100 Feasible Initially 6,734.3 149
PSO 100 Feasible Initially 15,157 2
IGWO 100 3 13,366 149
IGWO-1 100 3 13,632 89
26
IGWO-2 100 38 13,478 137
WOA 100 4 13,232 22
NNA 100 2 14,357 119
PSO 100 5 52,513 5
IGWO 100 18 43,867 145
IGWO-1 100 5 44,507 107
83
IGWO-2 100 67 45,219 149
WOA 100 3 48,575 110
NNA 100 79 49,665 143
PSO 100 28 104,807 28
IGWO 100 44 98,233 44
IGWO-1 100 89 96,215 89
160
IGWO-2 50 > 150 96,531 > 150
WOA 100 6 97,369 70
NNA 100 44 102,518 123

Table 5.  Quantitative assessment for algorithms. In the same experiment group, the black bold comments
mean advantageous indicators. Significant values are in [bold].

Regards to IGWO Algorithm

To evaluate the convergence speed of the IGWO algorithm, seven different algorithms, including PSO, GWO,
­WOA30, ­NNA11,31, two variations of IGWO algorithm (IGWO-1 and IGWO-2), and the algorithm in this paper
(IGWO), were employed in the case study. A population size of 20 and 150 iterations were set, the number of
iterations is enough to to make the function variation converge within 1%. The results are showed in Fig. 20.
IGWO-1, IGWO-2 and NNA failed to generate any feasible conflict-free solution set for the scale of 270
flights. On the other hand, the remaining algorithms successfully produced feasible solution sets. Among the
feasible, GWO exhibited the highest quality but had relatively slower convergence speed. Although PSO and
WOA generated the feasible with ideal iterations, their total working times of UFMCs significantly deviated
from that of the other algorithms, indicating inefficient operational performance. In contrast, IGWO showed
the fastest convergence speed.
Furthermore, in Table 5, quantitative analyses were conducted on four major indicators for handling the scales
of 12, 26, 83, 160, and 270 flights per day. In general, IGWO has got the most advantageous indicators (as shown
in bold black in the table). WOA is competitive in the convergence speed, but performs poorly in the working
time. In contrast, NNA seems to be adapted to small-scale problem.
The performance of GWO, IGWO-1 and IGWO-2 showed improvements compared to the PSO algorithm
mentioned in ­Reference23. In the case of 270 flights, the proposed IGWO algorithm achieved an 18.75% improve-
ment in convergence speed, which is consistent with the qualitative analysis. Simultaneously, it incurred a
relatively minor loss in operational efficiency, with only 1.76% more total working time compared to GWO.
Comparative analysis across different scales revealed that IGWO exhibited strong generalization performance
compared to the other algorithms. Meanwhile, IGWO-1 demonstrated potential advantages in solving medium-
scale (83 and 160 flight movements) scheduling problems, showing promise for generating more efficient guid-
ance trajectories.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 20

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 21.  Gantt chart for guidance tasks and charging decisions. The green diamond represents the
occurrence of a charging event.

Regards to vehicle assignment

According to the consultation ­notice32, the estimated number-k of UFMCs in this case study is 9. The results
for optimization of UFMC assignment are shown in Fig. 21. The results indicate that each UFMC is assigned
28 to 34 flight tasks, with a standard deviation of 582s in working time. The assignment inequality index is
only 0.015731, achieving fair vehicle assignment. Each UFMC only needs to be charged once per day, which is
consistent with the total working time of 170,463s mentioned in Section 5.5. When 9 UFMCs are required, the
average working time per vehicle is 18,940s, less than half of the safe endurance (q). In fact, for each vehicle, the
charging process shown in Fig. 21 needs to be performed only every two days. During the redundant charging
time and idle time of UFMCs, they can serve as backups for each other. Considering that Ezhou is a specialized
cargo airport, the guidance tasks at night at this airport are relatively active, which is related to the characteristics
of distribution of freight flight.

Conclusion
In this study focused on the integrated optimization of scheduling for UFMCs on airport surface, we proposed
the concept of GUs and a quantification method for guidance following spacing. We established an integrated
UFMCs scheduling model, consisting of trajectory planning and vehicle assignment, considering the demands of
precise connecting among guidance tasks and the vehicle charging. Additionally, considering the vehicle dynam-
ics, we designed a high-precision method of guidance trajectory deduction and a conflict prediction model based
on protected zones. We also developed an IGWO algorithm fitting the integer encoding.
Secondly, we validated the proposed approaches using the case of Huahu Airport in Ezhou. The IGWO
demonstrated significant improvements in generating conflict-free feasible solution sets and generalization
performance. The IGWO-1 algorithm showed potential in generating high-quality feasible solution sets for
medium-scale optimization of scheduling for UFMCs. 64.44% and 25% of guidance tasks successfully avoided
potential conflicts through fine-tuning timing and detour, sacrificing only 2.85% of efficiency to ensure safety. In
the vehicle assignment module, the assignment inequality index was as low as 0.015731, indicating the achieve-
ment of collaborative planning and the achievement of balanced allocation of guidance tasks.
Finally, future research will explore tactical optimization strategies for flight delays or unexpected incidents
encountered by UFMCs, aiming to enhance the robustness of algorithms and models.

Data availability
The dataset used and analyzed in the current study is available from the corresponding author on reasonable.

Received: 27 November 2023; Accepted: 4 April 2024

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 21

Vol.:(0123456789)
 www.nature.com/scientificreports/

References
1. CAAC. 2021. Roadmap for application of airport un-manned driving equipment (2021-2025); civil aviation order [2021] no. 46.
Tech. Rep., Civil Aviation Administration of China, Beijing, China.
2. Zong, F., He, Z., Zeng, M. & Yixuan, L. Dynamic lane changing trajectory planning for cav: A multi-agent model with path pre-
planning. Transportmetrica Bhttps://​doi.​org/​10.​1080/​21680​566.​2021.​19890​79 (2021).
3. Morris, R. et al. Planning, scheduling and monitoring for airport surface operations. In AAAI-16 Workshop on Planning for Hybrid
Systems (2016).
4. Sirigu, G., Battipede, M., Clarke, J.-P. & Gili, P. A fleet management algorithm for automatic taxi operations. In International
Conference on Research in Air Transportation—Doctoral Symposium (2016).
5. van Oosterom, S. & Mitici, M. Analyzing the impact of battery capacity and charging protocols on the dispatchment of electric
towing vehicles at a large airport. In AIAA AVIATION 2022 Forum, https://​doi.​org/​10.​2514/6.​2022-​3920 (2022).
6. He, D. (2018). Research on Scheduling of Ground Service Vehicles for Large Airport Airside Transition–Illustrated by The Case of
Ferry Bus. Master’s thesis, Beijing Jiao-tong University, Beijing, China.
7. CAAC. 2022. Rules governing the administration of air traffic management for civil aviation; ccar-93tm-r6. Tech. Rep., Civil Avia-
tion Administration of China.
8. Wang, S., Che, Y., Zhao, H. & Lim, A. Accurate tracking, collision detection, and optimal scheduling of airport ground support
equipment. IEEE Internet Things J. PP, 1–1. https://​doi.​org/​10.​1109/​JIOT.​2020.​30048​74 (2020).
9. Zhang, J., Chong, X., Wei, Y., Bi, Z. & Yu, Q. Optimization of apron support vehicle operation scheduling based on multi-layer
coding genetic algorithm. Appl. Sci. 12, 5279. https://​doi.​org/​10.​3390/​app12​105279 (2022).
10. Bao, D.-W., Zhou, J.-Y., Zhang, Z.-Q., Chen, Z. & Kang, D. Mixed fleet scheduling method for airport ground service vehicles
under the trend of electrification. J. Air Transp. Manag. 108, 102379. https://​doi.​org/​10.​1016/j.​jairt​raman.​2023.​102379 (2023).
11. Shang, D. Research on Optimal Scheduling Method of the Special Electric Vehicle Working on Parking Apron. Master’s thesis, Civil
Aviation University of China, Tianjin, China (2017).
12. Cheung, A., Ip, W., Lu, D. & Lai, C. An aircraft service scheduling model using genetic algorithms. J. Manuf. Technol. Manag. 16,
109–119. https://​doi.​org/​10.​1108/​17410​38051​05741​12 (2005).
13. Bazargan, M. Airline Operations and Scheduling 2nd edn. (China Civil Aviation Publishing House, 2020).
14. Diepen, G., Akker, J.M., Hoogeveen, J.A. & Smeltink, J.W. Using column generation for gate planning at Amsterdam Airport Schiphol.
Tech. Rep., Department of Information and Computing Sciences, Utrecht University (2007).
15. Kuhn, K. & Loth, S. Airport service vehicle scheduling. Air Traffic Control Q. 18, 63–83. https://d ​ oi.o
​ rg/1​ 0.2​ 514/a​ tcq.1​ 8.1.6​ 3 (2010).
16. Andreatta, G., Giovanni, L. & Monaci, M. A fast heuristic for airport ground-service equipment-and-staff allocation. Proc. Soc.
Behav. Sci.https://​doi.​org/​10.​1016/j.​sbspro.​2013.​12.​817 (2014).
17. Zhang, T., Ding, M. & Zuo, H. Airport ground movement disruption recovery via mixed-integer programming. J. Beijing Univ.
Aeronaut. Astronaut. 44, 2356 (2018).
18. Wang, Y., C. and Ren. A model of gate allocation for parallel multi-runway hybrid operation from the perspective of fuel-saving
and carbon emission reduction. J. Transp. Inform. Saf. 39: 144 (2021).
19. Li, N. et al. An empirical study on low emission taxiing path optimization of aircrafts on airport surfaces from the perspective of
reducing carbon emissions. Energies 12, 1649. https://​doi.​org/​10.​3390/​en120​91649 (2019).
20. Liu, J., Tang, X. & Zhang, Y. E. A. Optimization of airborne guidance velocity profile for required time of arrival. Aeronaut. Comput.
Tech. 51, 525 (2021).
21. Tang, Y., Hu, M. & Huang, R. E. A. Aircraft taxi routes planning based on free time windows and multi-agent for A-SMGCS. Acta
Aeronaut. Astronaut. Sinica 36, 1627 (2015).
22. Ji, X., Xu, Y., Pu, Y., Hao, J. & Qin, W. Risk prediction model of passenger car following behavior under truck movement interrup-
tion of two-lane high-way in mountainous area. J. Jilin Univ. Eng. Technol. Ed.https://​doi.​org/​10.​13229/j.​cnki.​jdxbg​xb202​20744
(2022).
23. Sirigu, G., Clarke, J.-P., Battipede, M. & Gili, P. Hybrid particle swarm optimization with parameter fixing: Application to automatic
taxi management. J. Air Transp. 28, 1–13. https://​doi.​org/​10.​2514/1.​D0172 (2019).
24. Zhang, T., Zhu, X., Li, J., Chen, H. & Li, Z. Research on conflict detection model for taxi-in process on the apron based on aircraft
wingtip keypoint detection. IET Intell. Transp. Syst. 17, 878. https://​doi.​org/​10.​1049/​itr2.​12314 (2022).
25. Li, H., Delage, E., Zhu, N., Pinedo, M. & Ma, S. Distributional robustness and inequity mitigation in disaster preparedness of
humanitarian operations. Manuf. Serv. Operations Manag.https://​doi.​org/​10.​1287/​msom.​2023.​1230 (2023).
26. Zhu, X., Tang, X. & Han, S. Aircraft initial taxiing route planning based on petri net and genetic algorithm. J. Southwest Jiaotong
Univ. 48, 565 (2013).
27. Zhang, Y. & Zhou, X. Modified grey wolf optimization algorithm for global optimization problems. J. Univ. Shanghai Sci.
Technol.https://​doi.​org/​10.​13255/j.​cnki.​jusst.​20200​331002 (2021).
28. Li, N., Hu, R., Qian, B., Jin, H. & Yu, N. Research on time-dependent vehicle routing problem with multiple time windows. J. Syst.
Simul. 34, 1775 https://​doi.​org/​10.​16182/j.​issn1​00473​1x.​joss.​21-​0244 (2022).
29. Zhu, X., Xu, C., Qu, J. & Su, T. Design and simulation-based evaluation of taxiway operation scheme for multi-runway airport
maneuvering area. J. Syst. Simul.https://​doi.​org/​10.​16182/j.​issn1​00473​1x.​joss.​21-​0789 (2022).
30. Zhou, K. et al. Research on solving flexible job shop scheduling problem based on improved GWO algorithm SS-GWO. Neural
Process. Lett.https://​doi.​org/​10.​1007/​s11063-​024-​11488-1 (2024).
31. Sadollah, A., Sayyaadi, H. & Yadav, A. A dynamic metaheuristic optimization model inspired by biological nervous systems: Neural
network algorithm. Appl. Soft Comput. 71, 747. https://​doi.​org/​10.​1016/j.​asoc.​2018.​07.​039 (2018).
32. CAAC. 2015. Guidelines for the configuration of specialized equipment for flight support in civil transport airport. Tech. Rep.
AC-139-CA-2015-01, Civil Aviation Administration of China, Beijing, China.

Acknowledgements
This work is supported by the Central University Basic Research Fund of China [grant number ZHMH2022-
008] and the National Key Research and Development Program of China [grant number 2022YFB2602004].

Author contributions
D.Y.: Software, Experiment, Data processing, Writing - Original draft. X.Z.: Conceptualization, Methodology,
Software. Y.Z.: Help writing. Q.Z.: Case of car-following, Idea.

Competing interests
The authors declare no competing interests.

Additional information
Correspondence and requests for materials should be addressed to X.Z.

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 22

Vol:.(1234567890)
www.nature.com/scientificreports/

Reprints and permissions information is available at www.nature.com/reprints.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.

© The Author(s) 2024

Scientific Reports | (2024) 14:8547 | https://doi.org/10.1038/s41598-024-58918-7 23

Vol.:(0123456789)