基于VBGMM-DCNN的列车卫星定位欺骗干扰检测方法

面向基于全球导航卫星系统的铁路列车定位实施欺骗干扰的主动检测，在卫星定位解算层次，运用深度学习建模学习方法的优势，提出一种基于变分贝叶斯高斯混合模型-深度卷积神经网络（variational Bayesian Gaussian mixture model-deep convolutional neural network, VBGMM-DCNN）的列车卫星定位欺骗干扰检测方法。该方法首先提取能够充分体现欺骗干扰对定位解算过程作用影响的卫星观测特征参数，构建干扰检测特征矢量；然后，采用VBGMM模型拟合经过预处理的特征向量的概率分布，得到二维概率密度图；最后，将概率密度图用于DCNN模型实施欺骗干扰的检测决策。结合现场实验所得运行场景数据，利用实验室搭建的欺骗干扰测试环境实施了干扰注入测试与检验，结果表明,欺骗干扰检测性能随着DCNN网络深度的增加而提升，相对于常规有监督决策方法F1值最高提升44.68%。基于VBGMM-DCNN的欺骗干扰检测能够适应测试验证中运用的列车运行特征及定位观测条件，所达到的检测性能优于对比算法。

Spoofing detection method for satellite-based train positioning based on VBGMM-DCNN

Aiming at the active detection of spoofing in railway train positioning based on global navigation satellite system (GNSS), a spoofing detection method for satellite-based train positioning based on variational Bayesian Gaussian mixture model-deep convolutional neural network (VBGMM-DCNN) is proposed by taking the advantages of deep learning (DL) at the GNSS navigation calculation stage. Firstly, feature parameters of satellite observation that can fully represent the spoofing effect on the positioning calculation process are extracted to construct the feature vector. Then, the VBGMM model is used to fit the probability distribution of the preprocessed feature vector and the two-dimensional probability density map is obtained. Finally, the probability density map is used in DCNN model for identifying the spoofing detection decision result. With the operation scene data obtained from the field experiment, the spoofing injection test and verification are carried out with a specific spoofing test environment built in the laboratory. The results illustrate that the enhanced spoofing detection performance could be achieved with an increasing DCNN network depth. The value of F1 increases by 44.68 % compared with the conventional supervised decision method. The spoofing detection based on VBGMM-DCNN can adapt to the train operation characteristics and positioning observation conditions used in the test verification, and the detection performance is better than the comparison algorithm.