湍流扩散环境中基于前馈神经网络的信源定位算法  

Source localization algorithm based on feed-forward neural network in turbulent diffusion environment

作　　者：张利成 鲍煦[1] 李静 林锋[2] 宋铁成[3] Zhang Licheng;Bao Xu;Li Jing;Lin Feng;Song Tiecheng(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212013,China;Zhejiang Institute of Freshwater Fisheries,Huzhou 313001,China;School of Information Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]江苏大学计算机科学与通信工程学院,镇江212013 [2]浙江省淡水水产研究所,湖州313001 [3]东南大学信息科学与工程学院,南京210096

出　　处：《东南大学学报（自然科学版）》2023年第2期370-376,共7页Journal of Southeast University：Natural Science Edition

基　　金：江苏省六大人才高峰资助项目(XYDXX-115);江苏省研究生科研创新计划资助项目(KYCX22-3661)。

摘　　要：为实现湍流扩散环境中的信源定位,提出了一种基于前馈神经网络(FNN)的定位方法.在FNN中联合交叉熵函数和均方误差函数作为损失函数,并在数据输入前增加了批量归一化层防止测试集过拟合.利用仿真得到的分子浓度数据对FNN模型进行训练,并使用测试集验证了所提方法的性能.仿真结果表明:在20 m仿真长度的湍流管道中,对于定位和初速度识别均在20次迭代内达到收敛;得到的平均定位误差为0.2 m,且误差在0.69 m范围内的概率达到90%,初速度识别的准确率达到95.4%.所提方法可以实现不同分子释放速度条件下的信源定位并识别发射分子的初速度,极大地减少了接收机的布置.In order to realize source localization in a turbulent diffusion environment,a localization method based on a feedforward neural network(FNN)was proposed.The cross entropy function and the mean square error function were combined as the loss function in FNN,and a batch normalization layer was added to prevent overfitting of the test set before data input.The FNN model was trained using molecular concentration data obtained from the simulation platform,and the performance of the proposed method was verified using the test set.The simulation results show that for both localization and initial velocity identification in a turbulent pipe of 20 m simulation length,convergence is reached within 20 iterations.Moreover,the average localization error obtained is 0.2 m and the probability of the error within the range of 0.69 m reaches 90%,and the accuracy of initial velocity identification reaches 95.4%.The proposed method can achieve source localization and identify the initial velocity of transmitting molecules at different molecule release speeds,greatly reducing the arrangement of receivers.

关 键 词：湍流扩散 分子浓度 前馈神经网络(FNN) 初速度识别 信源定位 

分 类 号：TN92[电子电信—通信与信息系统]