基于灰色-RBF神经网络的传播损耗模型训练  

PROPAGATION LOSS MODEL TRAINING ALGORITHM BASED ON GREY-RBF NEURAL NETWORK

作　　者：李丽娜[1] 梁德骕 马俊[1] 涂志[1] Li Li ’ na;Liang Desu;Ma Jun;Tu Zhi(College of Physics, Liaoning University, Shenyang 110036,Liaoning, China)

机构地区：[1]辽宁大学物理学院,辽宁沈阳110036

出　　处：《计算机应用与软件》2016年第8期136-140,182,共6页Computer Applications and Software

基　　金：国家自然科学基金青年项目(61403176);辽宁省教育厅科学技术研究项目(L2013003)

摘　　要：室内信号传播损耗模型是基于信号强度测距法的射频识别定位技术的关键。但因室内环境较为复杂且受到多径效应等因素影响,传统的基于经验的信号传播损耗模型环境适应性差,导致测距定位误差较大;而利用传统的神经网络进行传播损耗模型训练则存在所需训练样本过多、硬件采集工作量大等缺点。针对以上问题,提出在变密度采样模式下的基于灰色理论与RBF神经网络相结合的传播损耗模型训练方法。基于灰色理论,利用少量样本预测得到更多样本,并与部分原始样本共同重组样本数据进行RBF网络的训练,以构建传播损耗模型。实验结果表明,该方法可以利用较少的训练样本准确地建立室内信号传播损耗模型,可以很好地满足室内测距定位的精度要求,并可大大减少样本采集工作量。Indoor signal propagation loss model is the key to the radio frequency identification ( RFID) localisation technology based onreceived signal strength indicator (RSSI). Because of the rather complex indoor environment and the influence of multipath effect, traditionalempirical signal propagation loss model has poor environmental adaptability, and this leads to bigger localisation error in ranging. Besides thetraining of propagation loss model using traditional neural network has the disadvantages of too much training samples required and heavycollection workload in hardware. To overcome the problems mentioned above, we put forward the training method for propagation loss model invariable density sampling mode, which is based on the combination of grey theory and radial basis function ( RBF) neural network. Based ongrey theory, more training samples can be forecasted by using fewer samples, and they are used together with part of the original samples toreconstruct the sample data for RBF neural network training, so as to build the propagation loss model. Experimental results show that byusing the proposed method, it is able to build the indoor signal propagation loss model accurately with less training samples, which can wellmeet the precision requirement of the indoor localisation and greatly reduce the workload of sample collection as well.

关 键 词：室内定位 射频识别 传播损耗模型 径向基神经网络 灰色理论 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TN911.7[自动化与计算机技术—控制科学与工程]