基于多级神经网络的被动声定位算法研究

摘要

In order to solve the problems of precise mathematic model which is hard to establish, nonlinearity when solving the position equations and multi-array data fusion, this paper presented a passive acoustic localization algorithm based on multistage neural network.It obtained the location of sound source by the first stage RBF neural network, which might include invalid data eliminated by decision rule.The valid data entered the second stage RBF neural network, and got the higher precision of localization.Simulated the performance of the algorithm based on multi-stage neural network.The simulation results indicate that passive acoustic algorithm based on multi-stage neural network can improve the localization accuracy, positioning speed and robustness, and its performance is better than the algorithm based on single RBF neural network and the traditional algorithms.Even after individual sensors fail, it works well.%为了解决精确数学模型难以建立且求解位置方程时的非线性问题和多阵列数据融合问题,提出基于多级神经网络的被动声定位算法.该算法通过第一级RBF神经网络对声源进行初次定位,并剔除无效数据;再将有效数据输入第二级RBF神经网络,得到置信度更高的声源坐标.仿真结果表明,基于多级神经网络的被动声定位算法定位精度高、速度快、鲁棒性好,其定位性能优于单RBF神经网络和常规算法,甚至在个别传感器失效时,仍然能够取得较好的定位效果.