基于L-M贝叶斯正则化方法的BP神经网络在潜艇声纳部位自噪声预报中的应用

Submarine sonar self-noise forecast base on BP neural network and L-M Bayesian Regularization algorithm

基于L-M贝叶斯正则化方法使BP神经网络在推广能力、收敛速度和逼近精度上能够获得很大的提高。文中将BP神经网络和L-M贝叶斯正则化算法相结合用于潜艇声纳部位自噪声预报。分析了影响声纳部位自噪声的各种参数。利用潜艇声纳实测数据进行网络训练,训练好的神经网络可以对潜艇声纳部位自噪声进行精确预报。

L-M（Levenberg-Marquart） Bayesian Regularization algorithm is combined with Back-Propagation neural network to conduce to better generalization and faster speed of convergence,and higher learning accuracy can be acquired when multiple parameters and large patterns are inputted.In this paper, Bayesian Regularization algorithm and BP neural network were combined to forecast submarine sonar self-noise. All kinds of parameters that have function to submarine sonar self-noise were analyzed.Actual data were utilized to train BP neural network that can be used to accurately forecast submarine sonar self-noise.