基于遗传BP神经网络的海底沉积物声速预报  

A study on forecasting sound velocity of sea-floor sediments based on GA-BP method

作　　者：陈文景[1,2,3] 郭常升[1,2] 王景强[4] 侯正瑜[1,2,3] 

机构地区：[1]中国科学院海洋研究所,山东青岛266071 [2]中国科学院海洋地质与环境重点实验室,山东青岛266071 [3]中国科学院大学,北京100049 [4]国家海洋局第一海洋研究所海洋沉积与环境地质国家海洋局重点实验室,山东青岛266061

出　　处：《海洋学报》2016年第1期116-123,共8页

基　　金：海洋公益性行业科研专项项目(200905025)

摘　　要：在海底沉积物声速预报中,针对传统经验公式存在预测精度差、适用范围窄、缺乏物理意义等问题,在已有BP神经网络预测的基础上,运用遗传算法优化其初始权值和阈值的方法,构建出基于含水量、孔隙度的声速预报模型。将南沙海域采集得到的海底沉积物样品分为两部分,抽取120组涵盖陆架、陆坡、海槽等地貌单元的样品作为训练数据,另外剩余6组作为测试数据。经试验对比后发现,在对本区域进行声速预报时,宜采用遗传算法优化的BP神经网络,其要优于传统的单参数、双参数回归拟合预报方法和国内外其他学者所得到的经验公式。此种预报方法具有一定的科学依据和广泛的应用前景,可在今后为建立明确、统一的声速预报模型提供参考。In the sea-floor sediments velocity prediction,there exist many problems according to the empirical equations,such as poor accuracy,the narrow scope of application,lack of exact physical meaning.Based on the existing BP neural network,genetic algorithm（GA）is used to optimize the initial weights and threshold.A sea-floor sediment sound velocity forecasting model is established with the relationship of water content,porosity and velocity.Measurement data of study samples from the southern South China Sea are applied.These data are divided into two parts,120 groups including continental shelf,slope,trough samples selected as the training data,the other 6groups as test data.Experiments show that BP neural network based on GA is superior to the traditional single-parameter,double-parameter sound velocity forecasting empirical equation,which is recommended for the forecasting sound velocity of sea-floor sediments.This GA-BP method has certain scientific basis and broad application prospects in the future,can provide reference for the establishment of the accurate,uniform model.

关 键 词：遗传算法 BP神经网络 海底沉积物 声速预报 

分 类 号：P733.23[天文地球—物理海洋学]