A knowledge-transfer-based learning framework for airspace operation complexity evaluation. (October 2018)

• Record Type: Journal Article
• Authors: Cao, Xianbin; Zhu, Xi; Tian, Zhencai; Chen, Jun; Wu, Dapeng; Du, Wenbo
• Is Part Of: Transportation research. Volume 95(2018)
• Page Start: 61
• Page End: 81
• Publication Date: 2018-10
• Subjects: Airspace operation complexity -- Complexity evaluation -- Machine learning -- Small samples -- Knowledge transfer -- Evaluator integration; Transportation -- Periodicals; Transportation -- Technological innovations -- Periodicals; 388.011
• DOI: 10.1016/j.trc.2018.07.008
• Languages: English
• ISSNs: 0968-090X

Abstract

Highlights: Study airspace operation complexity evaluation with machine learning. Mappings between complexity level and its related factors are constructed. Proposed model is trained by the samples of both target and non-target airspace. Multi-factor-subset-generation and multi-base-evaluator-construction are employed. Discriminate evaluators built by non-target samples through local confidence analysis. Abstract: A sector is a component airspace whose operation is allocated to an air traffic controller. The operation complexity of a sector plays a critical role in the current Air Traffic Management system, e.g. it determines the workload volume of air traffic controllers and serves as a reliable index for airspace configuration and traffic flow management. Therefore, accurately evaluating the sector operation complexity is a problem of paramount importance in both practice and research. Due to numerous interacting factors, traditional methods based on only one single complexity indicator fail to accurately reflect the true complexity, especially when these factors are nonlinearly correlated. In light of these, the attempt to use machine learning models to mine the complex factor-complexity relationship has prevailed recently. The performance of these models however relies heavily on sufficient samples. The high cost of collecting ample data often results in a small training set, adversely impacting on the performance that these machine learning models can achieve. To overcome this problem, this paper for the first time proposes a new sector operation complexity evaluation framework based on knowledge transfer specifically for small-training-sample environment. The proposed framework is able to effectively mine knowledge hidden within the samples of the target sector, i.e. the sector undergoes evaluation, as well as other sectors, i.e. non-target sectors. Moreover, the framework can properly handle the integration between the knowledge derived from different sectors. Extensive experiments on real data of 6 sectors in China illustrate that our proposed framework can achieve promising performance on complexity evaluation when only a small training set of the target sector is available.