基于信号统计模型的变电站半遮挡融合定位方法

变电站存在建筑遮挡和电磁干扰等问题，这导致传统的基于电磁波定位的人员管控方法精度快速下滑。为避免因单传感器定位精度劣化而导致的电力安全管控效率降低问题，研究基于多源信息融合的巡检人员位置估计技术至关重要，而现有融合定位方法大多难以在地图信息未知的条件下鲁棒地选择传感器融合策略，因此文中提出一种基于卫星和近超声信号特征分析的融合定位方法，仅依靠信号统计特征实现环境信息判别并自适应选取融合策略。首先，利用多传感器信号特征统计模型构建指纹库，并基于t分布随机近邻嵌入(t-distributed stochastic neighbor embedding,t-SNE)降维算法和密度峰值聚类(density peaks clustering,DPC)算法处理指纹库数据。其次，依据聚类结果搭建反向传播(back propagation，BP)神经网络，将信号环境特征与卡尔曼滤波器的参数映射。最后，使用神经网络输出优化基于卡尔曼滤波的多源定位切换模型，形成自适应的融合定位方法。利用真实变电站半遮挡环境采集数据进行实验，结果表明，相较于未知环境信息、已知环境信息的融合定位方法，所提出的方法在地图信息未知的情况下节约了地图标定信息，实现了高鲁棒的位置估计。

Semi-occlusion substation fusion positioning method based on multi-sensor signal statistical model

There are problems such as building occlusion and electromagnetic interference in substations,which lead to a rapid decline in the accuracy of traditional personnel control methods based on electromagnetic wave positioning. In order to avoid the reduction of power safety management and control efficiency due to the degradation of single sensor positioning accuracy,it is very important to study the position estimation technology of patrol personnel based on multi-source information fusion. However,most of the existing fusion localization algorithms are difficult to robustly select sensor fusion strategies under the condition of unknown map information. A fusion positioning method based on satellite and near-ultrasonic signal feature analysis is proposed in this paper,which only relies on signal statistical features to realize environmental information discrimination and adaptively select fusion strategies. Firstly,the fingerprint database is constructed by using the multi-sensor signal feature statistical model,and the fingerprint library data is processed based on the t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction algorithm and the density peaks clustering (DPC) algorithm fingerprint library data. Secondly,a back propagation (BP) neural network is built based on the clustering results,and the signal environment features are mapped to the parameters of the Kalman filter. Finally,the neural network output is used to optimize the Kalman filter-based multi-source positioning switching model to form an adaptive fusion positioning algorithm. Experiments are carried out using data collected in a real substation semi-occluded environment. The results show that,compared with the fusion positioning method of unknown environmental information and known environmental information,the proposed method saves map marking information when the map information is unknown and realizes highly robust location estimation.