一种基于SO-CNN模型的可见光室内定位优化方法  

An Optimization Method for Visible Light Indoor Positioning Based on SO-CNN

作　　者：陈静[1] 刘旋 王金元[3] 章永龙 朱俊武[1] CHEN Jing;LIU Xuan;WANG Jinyuan;ZHANG Yonglong;ZHU Junwu(College of Information Engineering(College of Artificial Intelligence),Yangzhou University,Yangzhou 225127,China;School of Computer Science and Engineering,Southeast University,Nanjing 211189,China;Automotive New Technique of Anhui Province Engineering Technology Research Center,Anhui Polytechnic University,Wuhu 241000,China)

机构地区：[1]扬州大学信息工程学院(人工智能学院),江苏扬州225127 [2]东南大学计算机科学与工程学院,南京211189 [3]安徽工程大学汽车新技术安徽省工程技术研究中心,安徽芜湖241000

出　　处：《电讯技术》2024年第5期702-709,共8页Telecommunication Engineering

基　　金：江苏省“双创博士”项目(JSSCBS20211035);江苏省博士后基金项目(2021K402C);汽车新技术安徽省工程技术研究中心开放课题项目(QCKJ202205A)。

摘　　要：针对基于机器学习的可见光室内定位方法存在的手工调参、定位精度低等问题,结合蛇优化(Snake Optimization,SO)算法的寻优能力与卷积神经网络(Convolutional Neural Network,CNN)处理复杂非线性问题的能力,提出了一种基于SO-CNN模型的可见光室内定位优化方法。在考虑多径效应影响的情况下,采集每个位置点处的信噪比和对应位置坐标构建指纹数据库,对SO-CNN模型进行训练和测试,以得到最佳定位模型。实验结果表明,在5 m×5 m×3 m的房间中,与未经优化的CNN相比,该方法的平均定位误差降低了35.13%;与反向传播神经网络(Back Propagation Neural Network,BPNN)、多层感知器(Multilayer Perceptron,MLP)、SO-MLP相比,该方法的平均定位误差分别降低了54.75%,48.08%,37.01%。To solve the problems of manual parameter adjustment and low positioning accuracy in machine learning-based visible light indoor positioning(VLIP),VLIP optimization approach based on the Snake Optimization(SO)algorithm and Convolutional Neural Network(CNN)(SO-CNN)model is proposed,which combines the optimization-seeking capabilities of the SO algorithm with the ability of the CNN to handle complex nonlinear problems.Considering the influence of multipath effects,the fingerprint database is constructed by collecting the signal-to-noise ratio at each location point and the corresponding coordinate,and the SO-CNN model is trained and tested to obtain the best positioning model.Experimental results show that in a room of 5 m×5 m×3 m,the average positioning error of the proposed method is reduced by 35.13%compared with that of an unoptimized CNN,and by 54.75%,48.08%,and 37.01%,respectively,compared with that of Back Propagation Neural Network(BPNN),Multilayer Perceptron(MLP),and SO-MLP.

关 键 词：可见光室内定位(VLIP) 指纹定位法 蛇优化算法 卷积神经网络 

分 类 号：TN92[电子电信—通信与信息系统]