基于先验知识的多类CVM航班延误预警模型

Prior knowledge based multi-class core vector machine for flight delay early warning

基于使用现有的支持向量机解决机场航班延误预警问题存在未充分利用先验知识和训练需花费大量时间和空间的问题,提出了基于中心约束最小闭包球的加权多类算法。该算法首先利用先验知识确定一种新的基于相对紧密度的方法计算样本权值并将其融合到支持向量机中,然后转化为中心约束的最小闭包球进行训练。实验结果表明,该方法比现有的支持向量机具有更合理的分类面并且训练速度得到大大提高。

The early warning of airport runtime flight delay is a multi-class classification problem.There are two issues when solving this problem using the normal Support Vector Machine（SVM）.The first issue is that the prior knowledge is not adequately utilized,and the second issue is intensive time and space consumption for data training.A new algorithm,which is called as center-constrained Minimum Enclosing Ball（MEB）based weighted margin multi-class algorithm is proposed.First,the proposed algorithm uses the prior knowledge to build a new methodology which is based on a new relative affinity function.Then this new methodology is used to calculate the weights of the sample data and add them to the SVM.After adding these features,the SVM is converted to a center-constrained MEB and can be trained easily.Experiments show that the proposed algorithm not only gives more reasonable classification results comparing to normal SVM,but also obviously speeds up the data training processing.