基于改进支持向量机的空域交通态势识别方法

A Method for Monitoring Traffic State in the Airspace Based on an Improved Support Vector Machine

为了准确分析空域交通态势,针对当前空域交通态势感知中管制行为因素难以度量的问题,从多种管制员工作负荷的角度进行量化分析。借助全空域及机场模型软件(Total Airspace and Airport Modeller,TAAM)建立空域仿真模型,研究了栅格化空域场景下基于改进支持向量机(support vector machine,SVM)的空域交通态势识别方法。通过对比不同的栅格化方案,结合管制员实际运行经验确定栅格形状和尺寸,以边长为25 km的六边形为最小单元,对目标空域进行栅格化处理。在引入多种管制负荷的基础上同时考虑导航设施的分布情况,建立空域交通态势识别指标体系,运用K-means聚类算法对降维后的仿真样本数据进行聚类分析,获取先验分类数据。在支持向量机模型的基础上,引入麻雀搜索算法(sparrow search algorithm,SSA),构建基于麻雀搜索算法优化支持向量机(SSA-SVM)的空域交通态势识别模型。依据适应度对解集进行划分,对模型关键参数核函数参数σ和惩罚系数C进行优化,确定了1组泛化能力强同时避免过拟合问题的参数组合,并将栅格化空域交通态势划分为4个等级。以西安区域管制区为对象开展仿真实验,结果表明:与基于遗传算法优化支持向量机(GA-SVM)的空域交通态势识别模型相比,SSA-SVM模型克服了GA-SVM模型确定的的过拟合问题,平均分类识别准确率提高2.50%,最佳分类识别准确率提高1.73%;在176个栅格中,拥堵态、拥挤态和平稳态栅格个数分别为26、18和51,模型识别结果与基于管制员经验划分的复杂空域相比,覆盖率可达95%,验证了提出方法对空域交通态势识别及降低管制员工工作负荷的有效性。

This paper quantitatively analyzes the methods for monitoring airspace traffic state from the perspective of the workload of air traffic controllers,due to the difficult measurement of such a factor in current studies.In response to the need of monitoring airspace traffic state more efficiently,an airspace simulation model is developedbased onTotal Airspace and Airport Modeller(TAAM)software and a method for identifying traffic state in the rasterized airspace scenario is proposed based on an improved support vector machine(SVM).Based on real-world operation experience of controllers,the shape and size of grids are determined by comparing different rasterization schemes.Target airspace is rasterized by taking the hexagon with a side length of 25 km as the smallest unit.Considering a variety of controller's workloads and the distribution of navigation facilities,a set of indicators for describing traffic states of the airspace is developed.Ak-means clustering algorithm is used to generate prior classified data by aggregating simulated sample data.A traffic state model for the airspace,developed based on the sparrow-search algorithm(SSA)and SVM,iscalled SSA-SVM.The solution set is divided according to the fitness.Moreover,key parameters of the model,including kernel parametersσand penalty coefficients C,are optimized to determine a combination of parameters,which can increase the generalization capability of the model and avoid overfitting.Traffic states in the rasterized airspace are divided into four levels.Simulations are conductedfor the control airspace ofthe City of Xi'an.Study results show that the proposed SSA-SVM model can mitigate the overfitting problem,but not by the proposed genetic algorithm and support vector machine(GA-SVM)model.The average accuracy of classification is improved by 2.50%,and the accuracy of classification is improved by 1.73%.Among the tested 176 grids,the number of congested,crowded,and steady grids are 26,18,and 51,respectively.Compared with the partition method for the complex airspace based on controller experience,the convergencerate of the proposed model is as high as 95%,which verifies the effectiveness of the proposed method for identifying airspace traffic state and reducing the workload of air traffic controllers.