基于MIC-LSTM-ATT的机场动态容量评估研究  

Research on Airport Dynamic Capacity Prediction Based on MIC-LSTM-ATT

作　　者：李雅聪 邵荃[1] 唐明 沈志远[1] LI Ya-cong;SHAO Quan;TANG Ming;SHEN Zhi-yuan(Nanjing University of Aeronautics and Astronautics,Nanjing 211000,China;East China Regional Air Traffic Management Bureau of Civil Aviation of China,Shanghai 200000,China)

机构地区：[1]南京航空航天大学,江苏南京211000 [2]中国民用航空华东地区空中交通管理局,上海200000

出　　处：《航空计算技术》2023年第5期67-71,共5页Aeronautical Computing Technique

基　　金：国家自然科学基金委员会-中国民用航空局联合研究基金项目资助(U2233208)。

摘　　要：机场容量评估是空中交通管理和机场运行效率的关键,然而由于相关特征呈现非线性、时变性等复杂特性,机场动态容量评估仍具挑战。针对预测精度和可靠性问题,提出一种改进的深度学习机场动态容量评估方法(MIC-LSTM-Attention),主要包括基于最大互信息系数(MIC)进行相关性分析和引入注意力机制改进LSTM模型两部分。模型通过引入MIC确定强关联气象特征作为预测模型的输入,并应用注意力机制进行权重分配使得模型能够将注意力集中于重要气象信息上。实验数据为国内某大型国际机场一年内的航班运行及天气数据,实验建立了多个基线模型与所提模型进行对比分析。结果表明所构建的MIC-LSTM-ATT模型相较于其他3种方法拟合效果更好,准确度更高。Airport capacity prediction is the key to air traffic management and airport operation efficiency.Yet,airport dynamic capacity prediction is still a challenge for the nonlinear and time-varying characteristics of relevant features.To solve the prediction accuracy and reliability problems,an improved deep learning airport dynamic capacity prediction method(MIC-LSTM-Attention)was proposed,which mainly includes correlation analysis based on maximum information coefficient(MIC)and improved LSTM model integrating attention mechanism.The model used MIC to determine the strongly correlated meteorological features as the input to the prediction model,and used the attention mechanism to assign weights so that the model can focus its attention on important meteorological information.The experimental data are flight operation and weather data of a large international airport in China within one year.Multiple baseline models were established to compare and analyze the proposed models.The results show that the constructed MIC-LSTM-ATT model has better fitting effect and higher accuracy than the other three methods.

关 键 词：机场容量 气象因素 最大互信息系数 长短时记忆神经网络 注意力机制 

分 类 号：V355[航空宇航科学与技术—人机与环境工程]