MAML–SGD: a reliable airline rescheduling algorithm for small-sample learning based on MAML and SGD

• Published in: The Journal of supercomputing Vol. 80; no. 10; pp. 14953 - 14977
• Main Authors: Shen, Zhiyuan, Zhao, Qinshuai, Liu, Yiyang
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.07.2024; Springer Nature B.V
• Subjects: Aircraft; Airline scheduling; Airlines; Airports; Algorithms; Aviation; Compilers; Computer Science; Costs; Datasets; Flight plans; Interpreters; Machine learning; Multilayer perceptrons; Neural networks; Optimization; Parameter identification; Processor Architectures; Programming Languages; Rescheduling; Schedules; Airline rescheduling; MAML; Multilayer perceptron; Small-sample learning; Meta-learning
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-024-06014-y

Airline rescheduling can minimize the number of abnormal flights and their subsequent adverse effects during the flight-plan execution process. The current subjective manual rescheduling method has a low efficiency and does not consider uncontrollable factors. The traditional machine learning-based methods require massive training data, which are not applicable to some few sample's factors. In this study, we firstly addressed the factors that influence airline rescheduling. A set of airline rescheduling indicators was determined as the inputs of a multilayer perceptron model. Then, an airline rescheduling algorithm, namely MAML–SGD was proposed to solve the mentioned model constraint with few samples. Compared with the identification model that uses an SGD algorithm to iteratively update parameters, this proposed algorithm can maintain the stability of the gradient descent during the training and testing processes of the model. Finally, by using historical data from Guangzhou Baiyun International Airport and Urumqi Diwopu International Airport, the airline rescheduling accuracy of the model reached 97%, which was markedly higher than the results obtained by traditional machine learning models such as SVM.