基于多层感知机的远距离水下声源定位技术研究

Research on Long Range Underwater Source Location Based on Multi-layer Perceptron

针对传统匹配场处理方法存在水下声源远距离定位失准以及定位时间过长的问题,提出了一种基于多层感知机的水下声源被动定位方案。利用团队自研的“浮星”浮标实测的剖面数据模拟真实海洋环境,并在水中布设垂直水听器阵,仿真大量不同位置的声源在水听器处产生的接收数据。将多通道波形数据直接作为输入对多层感知机网络进行训练,从而获取高精度的定位神经网络。仿真结果表明,与匹配场处理算法相比,设计的感知机网络可以在大范围信噪比环境中实现有效的水下声源定位,其中在30dB信噪比下定位距离和深度的平均相对误差达到了1.94%和6.84%。另外,相对于失配声速剖面,使用平均声速剖面模拟的接收数据可提高网络的定位性能。

Aiming at the problems of inaccurate long range positioning of underwater source and long positioning time with traditional matched field processing(MFP)method,a long range location scheme of underwater source based on multi-layer perceptron(MLP)is proposed.The real marine environment is simulated by using the profile data measured by the team's self-developed“floating star”buoy.Vertical array of hydrophones is arranged in the water to simulate the received data generated by a large number of sources with different positions at the hydrophone.The multi-channel waveform data is directly used as the input to train the multi-layer perceptron,so as to obtain a high-precision positioning neural network.The simulation results show that compared with the matched field processing algorithm,the designed perceptron network can effectively locate the underwater source in a wide range of signal-to-noise ratio environment.The mean relative errors of location distance and depth reach 1.94%and 6.84%at 30dB signal-to-noise ratio.In addition,compared with the mismatched sound velocity profile,the received data simulated by the mean sound velocity profile can improve the positioning performance of the network.