LINC-BIT/AirCa

Contents

• 1. About Dataset
• 2. Download
• 3. Description
  • 3.1 AirCa-W
  • 3.2 AirCa-N
  • 3.3 Constraints description
• 4. Leaderboard
  • 4.1 Long-term Cargo Capacity Prediction
  • 4.2 Optimization of Cargo Loading
  • 4.3 Cargo balancing/loading with Large Language Model optimization
• 5. The AirCa APIs
• 6. References

Dataset Download: https://huggingface.co/datasets/LINC-BIT/AirCa
Dataset Website: https://huggingface.co/datasets/LINC-BIT/AirCa
Code Link: https://github.com/LINC-BIT/AirCa
Paper Link:

1. About Dataset

AirCa is a publicly available aircraft cargo loading dataset with millions of instances from industry. It has three unique characteristics: (1) Large-scale, AirCa contains in total 6,071k records and 1,092k flights, covering 6 aircraft types and 425 airports over a total span of 9 months. (2) Comprehensive information, AirCa is delivered to provide rich information pertaining to aircraft cargo loading, including detailed cargo characteristic information, loading-event logs, flight destination, and comprehensive loading constraints in practical scenarios. (3) Diversity, AirCa aims to increase data diversity from three perspectives: destination diversity, Flight diversity, and Constraint diversity.

The figure depicts the process of air cargo loading, starting with terminal administration, where goods are processed and prepared for transportation. Cargo loading follows as goods are transferred to the aircraft, and then flight preparation and flying take place as the plane gets ready for departure. The cargo is carefully organized in the Unit Load Devices (ULD), which are containers or pallets used to carry the cargo efficiently. For wide-body aircraft cargo holds, like the B777, there are designated areas for both small ULD containers and larger pallets. Meanwhile, narrow-body aircraft cargo holds, like the A320, have a different arrangement suited for smaller loads. The cargo types include bulk cargo and special goods, which require specific handling due to their size, fragility, or value.

2. Download

AirCa can be used for research purposes. Before you download the dataset, please read these terms. Then put the data into "./data/raw/".
The structure of "./data/raw/" should be like:

* ./data/raw/  
    * split_by_aircraft_type    
        * A320.csv   
        * ...    
    * split_by_date  
        * BAKFLGITH_LOADDATA2024-10-12.csv  
        * ...

import pandas as pd
>>> import pandas as pd
>>> df = pd.read_csv("BAKFLGITH_LOADDATA2024-10-12.csv")
>>> df.head(3)
       FLIGHT  TYPE DEST  WEIGHT  ... CONT PRIORITY VOLUME  SPECIAL CARGO
0  3744617311  A320  SIN     177  ...  NaN        1    0.0            NaN
1  3744617311  A320  SIN     177  ...  NaN        1    0.0            NaN
2  3744617332  A320  SIN     560  ...  NaN        1    0.0            NaN

3. Description

Below is the detailed field of each sub-dataset.

3.1 AirCa-W

Data field	Description	Unit/format
Cargo information
Loading order	Record of the cargo loading order	String
ID	Unique identifier for ULD	ID
Weight	Weight of ULD	String
ULD type	Types of ULD include general cargo, special cargo	String
Priority	Cargo loading priority	String
Length	Length of ULD	Float
Width	Width of ULD	Float
Height	Height of ULD	Float
Transship cargo	The record of whether it is transship cargo	Bool
Flight information
Loading time	Record of the cargo loading time	Time
Flight ID (anonymity)	Record of the different flights	ID
Destination airport	Record of the airport's name	String
Segment	The record of whether it is multi-segment flight	Bool
Aircraft information
Aircraft type	The type of the aircraft	String
Constraints	The constraints of air cargo loading	Constraint format

3.2 AirCa-N

Data field	Description	Unit/format
Cargo information
ID	Unique identifier for ULD	ID
Weight	Weight of cargo	String
Bulk type	Types of bulk include general cargo, special cargo	String
Priority	Cargo loading priority	String
Volume	The volume of the cargo	Float
Flight information
Loading time	Record of the cargo loading time	Time
Flight ID (anonymity)	Record of the different flights	ID
Destination airport	Record of the airport's name	String
Segment	The record of whether it is multi-segment flight	Bool
Aircraft information
Aircraft type	The type of the aircraft	String
Constraints	The constraints of air cargo loading	Constraint format

3.3 Constraints description

Constraint	Description	Unit/format
Cargo constraints
Special cargo space weight constraint	This constraint defines the maximum allowable weight of special cargo in the space	Float
Dangerous cargo isolation constraint	Any two special cargo loading locations need to maintain a specified distance	String
Aircraft cargo hold constraints
ULD correspondence constraint	Get the corresponding relationship of cargo types and verify each piece of cargo data	String
ULD Type Restriction Rules	If the container type in the loading data is not one of the ones defined in ULD Type, the check fails	Bool
ULD type and ULD number constraint	If the container type does not correspond to the container serial number, the verification fails	Bool
Cargo hold availability constraint	Before loading, check whether the cargo hold is available	String
Mixed cargo space constraint	Check whether there is mixed loading in the cargo hold	Bool
Number of ULD constraint	The quantity of ULD cannot exceed this specified value	Float
Front/Rear compartment constraint	Ensure weight in the front (FWD) and rear (AFT) compartments do not exceed the defined limits.	Float
Cargo Type validity constraint	Check whether cargo type is valid and belongs to predefined cargo types.	String
Loading constraints
Weight constraint	Maximum load weight of the cargo hold	Float
CG constraint	Ideal center of gravity range for airliner when zero fuel	Float
Volume constraint	The volume of cargo cannot exceed this specified value	Float
Joint weight constraint	Total load weight constraints for multiple cargo holds	Float
Cargo space weight constraint	This constraint defines the maximum weight limit for a cargo space.	Float
Continuous loading constraint	Some types of ULDs need to be loaded according to the load sequence	String
Load order constraint	Goods must be loaded in the specified order	String

4. Leaderboard

Blow shows the performance of different methods in AirCa.

4.1 Long-term Cargo Capacity Prediction

4.2 Optimization of Cargo Loading

Experimental results of Optimization of Cargo Loading. The introduction of 12 baselines is shown as follows:

• COM [1]: Combinatorial Optimization Model solves discrete optimization tasks by searching for an optimal arrangement among a finite set of feasible solutions.
• IOM [2]: Improved Combinatorial Optimization Model obtains better solutions for discrete optimization tasks by refining search strategies to more effectively explore feasible configurations.
• NL-CPLEX [3]: NL-CPLEX addresses nonlinear optimization tasks by leveraging branch-and-bound and cutting-plane techniques to efficiently explore the solution space.
• SDCCLPM [4]: Stochastic-Demand Cargo Container Loading Plan Model optimizes container loading configurations under demand uncertainty by incorporating probabilistic approaches to balance capacity and cost requirements.
• MLIP [5]: Mixed Integer Linear Program finds optimal solutions to discrete optimization problems by combining integer constraints with linear relationships in a branch-and-bound search process.
• MLIP-WBP [6]: MLIP-WBP optimizes weighted bin packing by employing a Mixed Integer Linear Programming formulation to balance item distribution and capacity constraints.
• MLIP-ACLPDD [7]: MLIP-ACLPDD solves advanced cargo loading planning under uncertain demand by incorporating robust constraints into a Mixed Integer Linear Programming framework.
• HGA [8]: Hybrid Genetic Algorithm enhances solution quality by combining evolutionary operators with complementary search techniques to accelerate convergence and explore the solution space more thoroughly.
• GA-normal [9]: GA-normal employs foundational genetic algorithm operations—selection, crossover, and mutation—to explore solutions within a population-based search framework.
• DMOPSO [10]: Discrete Multi-Objective Particle Swarm Optimization locates Pareto-optimal solutions in discrete search spaces by adapting swarm-based velocity and position update mechanisms to address multiple conflicting objectives.
• PSO-normal [11]: PSO-normal employs the basic velocity and position update rules, guided by personal and global best solutions, to iteratively converge on an optimal search space configuration.
• RCH [12]: Randomized Constructive Heuristic incrementally constructs feasible solutions by integrating stochastic choices during each step, thus diversifying the search process and enhancing solution discovery.

Method	B777 MAC(%)↓	B777 INDEX(%)↓	B777 TIME(s)↓	A320 MAC(%)↓	A320 INDEX(%)↓	A320 TIME(s)↓	B787 MAC(%)↓	B787 INDEX(%)↓	B787 TIME(s)↓
COM	23.93 ± 0.59	3.40 ± 1.64	0.06 ± 0.04	21.14 ± 0.28	6.46 ± 2.20	0.06 ± 0.05	23.71 ± 0.47	3.10 ± 1.58	0.03 ± 0.03
IOM	23.90 ± 0.59	3.40 ± 1.62	0.07 ± 0.08	21.16 ± 0.28	6.50 ± 2.16	0.07 ± 0.05	23.71 ± 0.46	3.08 ± 1.56	0.06 ± 0.05
NL-CPLEX	23.92 ± 0.58	3.45 ± 1.60	0.08 ± 0.06	21.15 ± 0.29	6.48 ± 2.18	0.08 ± 0.07	23.70 ± 0.47	3.07 ± 1.61	0.05 ± 0.04
SDCCLPM	23.91 ± 0.59	3.40 ± 1.63	0.07 ± 0.05	21.15 ± 0.28	6.46 ± 2.18	0.07 ± 0.06	23.70 ± 0.46	3.08 ± 1.57	0.05 ± 0.04
MLIP	23.92 ± 0.57	3.47 ± 1.59	0.06 ± 0.07	21.14 ± 0.29	6.45 ± 2.20	0.06 ± 0.05	23.69 ± 0.46	3.04 ± 1.63	0.03 ± 0.02
MLIP-WBP	23.92 ± 0.58	3.45 ± 1.60	3.53 ± 5.78	21.15 ± 0.29	6.47 ± 2.19	1.43 ± 0.78	23.70 ± 0.47	3.07 ± 1.61	1.43 ± 0.85
MLIP-ACLPDD	23.93 ± 0.59	3.44 ± 1.65	3.46 ± 1.61	21.14 ± 0.29	6.44 ± 2.20	1.46 ± 0.98	23.71 ± 0.47	3.12 ± 1.60	1.67 ± 1.02
HGA	23.37 ± 0.47	3.23 ± 1.06	253.30 ± 0.80	21.14 ± 0.22	6.69 ± 1.80	1.80 ± 0.84	23.46 ± 0.24	3.86 ± 1.74	193.62 ± 0.51
GA-normal	23.35 ± 0.48	3.13 ± 1.08	221.82 ± 0.52	21.14 ± 0.22	6.71 ± 1.80	1.81 ± 0.51	23.44 ± 0.23	3.73 ± 1.69	145.70 ± 0.17
DMOPSO	23.12 ± 0.49	1.56 ± 1.65	266.11 ± 2.61	21.10 ± 0.28	6.59 ± 2.43	2.60 ± 0.61	23.29 ± 0.29	3.00 ± 2.39	204.13 ± 2.02
PSO-normal	23.19 ± 0.44	2.13 ± 1.81	211.73 ± 2.70	21.09 ± 0.28	6.56 ± 2.43	2.61 ± 0.70	23.30 ± 0.27	3.09 ± 2.19	199.24 ± 1.80
RCH	23.35 ± 0.50	3.23 ± 1.23	200.63 ± 0.06	21.07 ± 0.24	6.55 ± 1.93	1.78 ± 0.06	23.41 ± 0.26	3.50 ± 1.93	200.20 ± 0.02

Method	Segment 1 MAC(%)↓	Segment 1 INDEX(%)↓	Segment 1 TIME(s)↓	Segment 2 MAC(%)↓	Segment 2 INDEX(%)↓	Segment 2 TIME(s)↓
COM	23.59 ± 0.40	2.72 ± 1.56	0.73 ± 0.61	24.29 ± 0.74	3.89 ± 2.32	1.25 ± 0.92
IOM	23.65 ± 0.41	3.02 ± 1.62	1.19 ± 0.90	24.30 ± 0.73	4.02 ± 2.42	1.82 ± 1.19
NL-CPLEX	23.61 ± 0.41	2.65 ± 1.49	1.06 ± 0.94	24.30 ± 0.74	3.88 ± 2.27	1.96 ± 1.39
SDCCLPM	23.63 ± 0.41	2.96 ± 1.61	1.11 ± 0.95	24.28 ± 0.74	3.97 ± 2.38	1.81 ± 1.35
MLIP	23.63 ± 0.42	2.68 ± 1.48	0.84 ± 0.75	24.28 ± 0.74	3.87 ± 2.24	1.21 ± 0.85
MLIP-WBP	23.61 ± 0.41	2.65 ± 1.49	32.06 ± 22.02	24.30 ± 0.74	3.88 ± 2.27	44.32 ± 22.77
MLIP-ACLPDD	23.60 ± 0.40	2.73 ± 1.54	34.05 ± 22.46	24.28 ± 0.74	3.88 ± 2.33	51.55 ± 27.58
HGA	23.44 ± 0.23	3.73 ± 1.69	36.10 ± 8.55	23.39 ± 0.30	3.32 ± 2.15	23.86 ± 2.69
GA-normal	23.43 ± 0.24	3.65 ± 1.77	28.70 ± 3.45	23.25 ± 0.24	2.37 ± 1.74	23.57 ± 2.50
DMOPSO	23.30 ± 0.27	2.58 ± 2.19	38.12 ± 23.18	23.20 ± 0.27	2.24 ± 2.25	37.39 ± 20.79
PSO-normal	23.29 ± 0.28	2.54 ± 2.01	67.54 ± 57.82	23.34 ± 0.27	3.43 ± 2.25	31.77 ± 16.63
RCH	23.39 ± 0.27	3.38 ± 1.96	36.72 ± 0.55	23.25 ± 0.27	2.35 ± 1.96	35.27 ± 0.31

4.3 Cargo balancing/loading with Large Language Model optimization

Method	B777 MAC(%)↓	B777 INDEX(%)↓	B777 TIME(s)↓	B787 MAC(%)↓	B787 INDEX(%)↓	B787 TIME(s)↓
HGA	23.44 ± 0.22	3.75 ± 1.65	4.91 ± 2.08	23.44 ± 0.21	3.73 ± 1.55	2.50 ± 1.12
GA-normal	23.43 ± 0.23	3.66 ± 1.67	2.39 ± 0.76	23.46 ± 0.21	3.89 ± 1.58	1.29 ± 0.17
DMOPSO	23.32 ± 0.28	3.21 ± 2.30	3.28 ± 2.23	23.39 ± 0.26	3.79 ± 2.10	1.60 ± 0.81
PSO-normal	23.39 ± 0.28	3.79 ± 2.30	5.80 ± 4.09	23.39 ± 0.29	3.78 ± 2.37	1.19 ± 0.74
RCH	23.39 ± 0.26	3.40 ± 1.91	3.66 ± 0.03	23.42 ± 0.24	3.61 ± 1.76	0.72 ± 0.01

5. The AirCa APIs

In addition to our AirCa dataset, we release the AirCa package, including three types of APIs. It is designed to faciliate researchers in developing aircraft cargo loading applications.The details are presented as follows:

DataDownloader. This API allows researchers to download the AirCa data. the code presents how to utilize the DataDownloader API to download the up-to-date AirCa data. DataDownloader. This API allows researchers to download the AirCa data. Figure 4 presents how to utilize the DataDownloader API to download the up-to-date AirCa data.

from api . download_airca import AirCaDownloader
downloader = AirCaDownloader ()
# Download data A320
downloader . download_AirCa ( url , path , aircraft_type =" A320 " ,
date =" 2024 -10 -12 ")
# Download data B737
downloader . download_AirCa ( url , path , aircraft_type =" B737 " ,
date = None )
# Download data for all available aircraft types
downloader . download_AirCa ( url , path , aircraft_type = None ,
date = None )

DataRetriever. This API enables researchers to conveniently obtain the AirCa data stroed in the local machine. For instance, the code shows how to employ the DataRetriever API to obtain the AirCa data for aircraft type B777.

from api . retriever import Retriever
retriever = Retriever ()
# Enter the type A320
retriever . retrieve ( path = path , aircraft_type = " A320 ")
# Enter the type B777
retriever . retrieve ( path = path , aircraft_type = " B777 ")
# Enter the type B787
retriever . retrieve ( path = path , aircraft_type = " B787 ")

DataLoader. This API is designed to assist researchers in their applications of aircraft cargo loading. It allows researchers to flex- ibly and seamlessly merge multiple modalities of AirCa data. It exposes the AirCa through a DataLoader object after performing necessary data preprocessing techniques. A PyTorch example of using our DataLoader API for training DNNs is shown in the code.

import torch
from torch . utils . data import DataLoader
# generate AirCa ( A320 ) dataset for training
dataloader1 = DataLoader ( AircraftDataset ( path ," A320 ") ,
batch_size = batch_size , shuffle = True )
# generate AirCa ( B777 ) dataset for training
dataloader2 = DataLoader ( AircraftDataset ( path ," B777 ") ,
batch_size = batch_size , shuffle = True )
# generate AirCa ( B787 ) dataset for training
dataloader3 = DataLoader ( AircraftDataset ( path ," B787 ") ,
batch_size = batch_size , shuffle = True )
train_model ( dataloader1 , baseline_name , criterion ,
optimizer , epochs =600)

6. References

[1] Zhao, X., Dong, Y., & Zuo, L. (2023). A combinatorial optimization approach for air cargo palletization and aircraft loading. Mathematics, 11(13), 2798.
[2] Mesquita, A. C. P., & Sanches, C. A. A. (2024). Air cargo load and route planning in pickup and delivery operations. Expert Systems with Applications, 249, 123711.
[3] Yan, S., Lo, C.-T., & Shih, Y.-L. (2006). Cargo container loading plan model and solution method for international air express carriers. Transportation Planning and Technology, 29(6), 445–470.
[4] Yan, S., Shih, Y.-L., & Shiao, F.-Y. (2008). Optimal cargo container loading plans under stochastic demands for air express carriers. Transportation Research Part E: Logistics and Transportation Review, 44(3), 555–575.
[5] Limbourg, S., Schyns, M., & Laporte, G. (2012). Automatic aircraft cargo load planning. Journal of the Operational Research Society, 63(9), 1271–1283.
[6] Zhao, X., Yuan, Y., Dong, Y., & Zhao, R. (2021). Optimization approach to the aircraft weight and balance problem with the centre of gravity envelope constraints. IET Intelligent Transport Systems, 15(10), 1269–1286.
[7] Lurkin, V., & Schyns, M. (2015). The airline container loading problem with pickup and delivery. European Journal of Operational Research, 244(3), 955–965.
[8] Zhu, L., Wu, Y., Smith, H., & Luo, J. (2023). Optimisation of containerised air cargo forwarding plans considering a hub consolidation process with cargo loading. Journal of the Operational Research Society, 74(3), 777–796.
[9] Chenguang, Y., Liu, H., & Yuan, G. (2018). Load planning of transport aircraft based on hybrid genetic algorithm. In MATEC’18, Vol. 179 (pp. 01007). EDP Sciences.
[10] Dahmani, N., & Krichen, S. (2016). Solving a load balancing problem with a multi-objective particle swarm optimisation approach: application to aircraft cargo transportation. International Journal of Operational Research, 27(1-2), 62–84.
[11] Dahmani, N., & Krichen, S. (2013). On solving the bi-objective aircraft cargo loading problem. In ICMSAO’13 (pp. 1–6). IEEE.
[12] Gajda, M., Trivella, A., Mansini, R., & Pisinger, D. (2022). An optimization approach for a complex real-life container loading problem. Omega, 107, 102559.