摘要
在声呐、雷达等设备的目标探测中,声源方位估计是需要解决的关键问题之一。针对水下传感器阵列接收信号的波达方向角(DOA)估计算法中,传统的BP神经网络算法会因网络参数不合理和层数过多导致过拟合的问题,以往通过粒子群算法(PSO)进行优化后,网络仍容易过早结束训练而导致性能不佳。为此,本文提出一种基于变分模态分解结合粒子群算法优化后的BP神经网络算法。首先对目标回波信号进行可变模态分解,对分解得到的各分量进行时频分析后叠加的谱图特征作为经粒子群算法优化后的BP神经网络算法的输入进行训练测试,以此来提高阵元接收目标回波的DOA估计精度。仿真实验结果表明,结合变分模态分解及粒子群算法优化的BP神经网络具有更好的识别效果和泛化能力,提高了DOA的估计精度。
In target detection of sonar,radar and other equipment,the estimation of sound source azimuth is one of the key problems to be solved.In the DOA estimation algorithm of received underwater sensor array signal,the triditional BP neural network algorithm will lead to the problem of over fitting due to unreasonable network parameters and too many layers.In the past,after optimization by particle swarm optimization(PSO),the network is still easy to end training too early,resulting in poor performance.Therefore,this paper proposes a BP neural network algorithm based on the combination of the variational mode decomposition and particle swarm optimization.Firstly,the echo signal of the target is decomposed by the variable mode,and the spectral characteristics of the decomposed components after the time-frequency analysis are used as the input of the BP neural network algorithm optimized by the particle swarm optimization for training and testing,so as to improve the receiving target echo of the array’s DOA estimation accuracy.The simulation results show that the BP neural network combined with the variational mode decomposition and particle swarm optimization has better recognition effect and generalization ability,and improves the estimation accuracy of DOA.
作者
杨阳
毛家林
潘成胜
YANG Yang;MAO Jia-lin;PAN Cheng-sheng(Key Laboratory of Communication and Network,Dalian University,Dalian 116622,China;College of Information Engineering,Dalian University,Dalian 116622,Chinan)
出处
《舰船科学技术》
北大核心
2021年第7期122-126,共5页
Ship Science and Technology
基金
国家自然基金资助项目(61540024)
国家自然基金资助项目(61901079)。
关键词
声呐目标DOA估计
变分模态分解
粒子群算法
BP神经网络
DOA estimation of sonar target
variational mode decomposition
particle swarm optimization
BP neural network
