卷积-长短期记忆神经网络超宽带定位方法

Convolution-long short-term memory neural network UWB localization method

针对室内视距环境下超宽带(UWB)观测值中的测距误差影响定位精度的问题,提出一种基于卷积神经网络与长短期记忆网络(CNN-LSTM)相结合的UWB测距误差改正模型:将基站与标签之间的测距值和Chan算法解算的标签初始坐标作为卷积神经网络(CNN)的输入,借助CNN良好的数据特征提取能力,充分挖掘UWB测距值的特征;然后利用长短期记忆网络(LSTM)进行进一步的特征学习,并进行训练和预测UWB测距值,以减少测距误差对UWB测距值精度的影响;最后,利用高斯-牛顿迭代算法求解出最终的UWB定位结果,同时,建立多项式和指数函数UWB测距误差改正模型,并与本文方法进行对比分析。实验结果表明,在静态和动态实验下,基于CNN-LSTM网络模型结果的精度均优于其他2种模型,证明该算法可有效降低测距误差,提高UWB的定位精度。

Aiming at the problem that the ranging error in the ultra wide band(UWB)observation value in the indoor line-of-sight environment affects the positioning accuracy,the paper proposed an UWB ranging error correction model based on the combination of convolutional neural network and long short-term memory network(CNN-LSTM):the ranging value between the base station and the tag and the initial coordinates of the tag calculated by the Chan algorithm were used as the input of the convolutional neural network(CNN),and with the help of CNN,which has good capabilities of data feature extraction,the characteristics of UWB ranging value were fully exploited;then the long short-term memory network(LSTM)was used for further feature learning and training and predicting the UWB ranging value to reduce the impact of ranging errors on the accuracy of the UWB ranging value;finally,the Gauss-Newton iterative algorithm was used to solve the final UWB positioning result,at the same time,a polynomial and exponential function UWB ranging error correction model was established and compared with the method proposed in this paper.Experimental results showed that in the static and dynamic experiments,the accuracy of the results based on the CNN-LSTM network model would be better than that of the other two models,indicating that the proposed algorithm could effectively reduce the ranging error and improve the indoor positioning accuracy of UWB.