滑行时间是表征机场场面运行效率和地面延误程度的重要指标，因此准确预测并优化滑行时间是提高场面运行效率的重要手段。在数据匮乏时期，滑行时间预测主要采用浅层结构算法，难以表达影响因子的复杂非线性作用，且建立在有限统计样本基础上的预测模型普适性较弱；而滑行时间优化缺少对场面交通流变化等时变因素的自适应调整，不能满足复杂多变环境的优化需求。因此，研究大数据环境下的滑行时间预测及优化关键技术尤为重要。本项目旨在基于海量ADS-B数据提取长时间序列滑行时间样本，分析滑行时间影响因子与典型时空模式；结合与滑行过程关联的环境参数数据，利用深度学习算法识别滑行时间预测的最优特征空间，构建滑行时间快速预测模型；并借助ADS-B数据构建融合反馈机制的多智能体系统，动态调整滑行路径进而实时优化滑行时间。研究成果将为运控及保障部门提供精准的起降相关信息，进而提高场面运行效率，减少地面延误及由此带来的经济损失。

Taxi time is a typical representative of airport surface operation efficiency and severity of flight ground delay. How to predict taxi time precisely and optimize taxi time is vital measures to improve surface operation efficiency. During the era of poor data, taxi time prediction usually takes simple structure algorithms such as regression analysis, which make it hard to describe the complicated non-linear effect of impact factors. In addition, as the samples is greatly limited, prediction model just only meet partial demands. Meanwhile, taxi time optimization always ignore dynamic surface traffic flow parameter. Based on these problems, this project aims to realize: (1) Extraction of taxi time for airports of different categories based on long time series of national ADS-B data. Then the determinants and typical spatio-temporal pattern will be worked out. (2) Modelling of taxi time quick prediction. This will adopt deep learning algorithm to identify optimal feature space for prediction based on the analysis of all relevant factors. (3) Real time adjust of taxi route to reduce taxi time through feedback-embedded multiple intelligent objects system. The achievements of this projects will provide accurate approach and departure information and taxi solution for airport ATC and ground support department. It will be of great help for follow-in of ground support tasks. It can also significantly improve surface operation efficiency and further reduce flight ground delay and economic loss.