摘要
针对在利用深度学习网络进行位置指纹定位时,模型参数值大多根据经验给出的问题,该文利用粒子群算法(PSO)对地磁数据分解和深度学习网络模型的参数进行优化,以得到最优定位模型。首先对5G信道状态信息(CSI)幅值和地磁数据进行降噪处理以提升指纹库质量,然后采用PSO算法对模型参数自动寻优,构建离线训练的多输入神经网络模型,最后利用实时采集到的传感器数据进行定位。实验表明,在会议室、教学楼大厅及IPIN2023_T7数据集3种场景下的平均绝对定位误差分别为1.1、2.13和1.65 m,较未进行参数优化的多输入神经网络模型算法分别提升了8.3%、9%和10.3%,具有更佳的定位性能和更好的泛化能力。
Aiming at the problem of using deep learning networks for location fingerprint localization,the selection of model parameters values are mostly given based on experience,swarm optimization(PSO)algorithm was used to optimize the parameters of geomagnetic data decomposition and deep learning network model in this paper,in order to obtain the optimal positioning model.Firstly,the amplitude of 5G channel state information(CSI)and geomagnetic data were denoised to improve the quality of the fingerprint database.Then,the PSO algorithm was used to automatically optimize the model parameters and an offline trained multi-input convolutional neural networks(CNN)model was constructed.Finally,real-time sensor data was used for localization.Experimental results showed that the average absolute positioning errors in the meeting room,teaching building hall and IPIN2023_T7 data set were 1.1,2.13 and 1.65 m,respectively,which were 8.3%,9%and 10.3%higher than the multi-input neural network model algorithm without parameter optimization,and had better positioning performance and generalization ability.
作者
杜莹
程振豪
孙丹丹
DU Ying;CHENG Zhenhao;SUN Dandan(Zhengzhou Normal University,Zhengzhou 450044,China;Information Engineering University,Zhengzhou 450001,China;Henan Jinyuan Construction Company,Zhengzhou 450001,China)
出处
《测绘科学》
北大核心
2025年第6期55-63,共9页
Science of Surveying and Mapping
关键词
指纹定位
地磁数据
信道状态信息
深度学习
粒子群算法
fingerprint localization
geomagnetic data
channel state information(CSI)
deep learning
particle swarm optimization(PSO)
