基于时空特征融合的飞机尾涡识别

为了实现动态尾流缩减技术,减少进近阶段前机尾流对后机飞行安全的影响。依据相干激光雷达（coherent Light Lidar,简称CDL）扫描风场循环周期性特点,提出一种基于时空特征融合的飞机尾涡识别模型。首先,CDL扫描生成的径向速度风场转换成序列输入和块输入。然后,双向长短时记忆（bidirectional long short-term memory, 简称Bi-LSTM）网络用于提取序列输入的时间特征,卷积神经网络（convolutional neural network, 简称CNN）网络用于提取径向速度风场块输入的空间特征。最后,将融合的时域和空域特征输入全连接层分类器,得到最终分类识别结果。实验团队在深圳宝安机场附近采集风场,并构建尾流数据集来验证所提得融合模型,结果表明：基于CNN和Bi-LSTM时空特征混合模型具有较好的分类性能,在尾涡识别上的准确率、召回率、F1分数分别达到97.13%、97.50%、97.03%,且相比单一模型是一种更有效的识别方式,能够获得实时高效尾流预警。

Aircraft wake vortex identification based on spatiotemporal feature fusion

In order to realize the dynamic wake reduction technology, the influence of the wake of the front aircraft on the flight safety of the rear aircraft during the approach phase is reduced. According to the periodic characteristics of coherent light lidar (CDL) scanning wind field, an aircraft wake vortex identification model based on fusion of spatiotemporal features is proposed. First, the radial velocity wind field generated by the CDL scan is converted into sequence input and patch input. Then, a bidirectional long short-term memory (Bi-LSTM) network is used to extract the temporal features of the sequence input, and a convolutional neural network (CNN) network is used to extract the radial velocity wind field block input spatial characteristics. Finally, the fused temporal and spatial features are input into the fully connected layer classifier to obtain the final classification and recognition result. The experimental team collected wind fields near Shenzhen Bao"an Airport, and constructed a wake data set to verify the proposed fusion model. The results show that the spatiotemporal feature mixture model based on CNN and Bi-LSTM has better classification performance, and it can be used in wake vortex recognition. The accuracy rate, recall rate, and F1 score reached 97.13%, 97.50%, and 97.03% respectively, and compared with a single model, it is a more effective identification method and can obtain real-time and efficient wake early warning.