Abstract

Based on the data of the sound speed and the physical–mechanical properties of 300 seafloor sediment samples collected in the middle area of the Southern Yellow Sea, a backpropagation (BP) neural network algorithm is used herein to study the neural network topology and the corresponding neuron characteristic factors. A neural network model for predicting the sound speed of the seafloor sediment based on density, water content, porosity, void ratio, liquid limit, plastic limit, and clay content in the middle area of the Southern Yellow Sea is established. The results show that the sound speed prediction accuracy is the highest when the network model contains seven input layer neurons, ten hidden layer neurons, and one output node. According to the prediction results, the MAE(Mean Absolute Error) and the MAPE(Mean Absolute Percentage Error) are 13.19 m s⁻¹ and 0.85%, respectively. The prediction results illustrate that the established BP neural network prediction method for the seafloor sediment sound speed is better than the traditional single, two and multi-parameter prediction equations.

摘要

本文基于南黄海中部地区采集的300个海底沉积物样品的声速和物理力学特性数据，采用反向传播（BP）神经网络算法研究神经网络拓扑结构和对应的神经元特征因子。建立了基于密度、含水量、孔隙度、空隙率、液限、塑限和黏土含量的南黄海中部海底沉积物声速预测神经网络模型。结果表明，当网络模型包含7个输入层神经元、10个隐藏层神经元和1个输出节点时，声速预测精度最高。根据预测结果，MAE（平均绝对误差）和 MAPE（平均绝对百分比误差）为 13。^分别为-1和 0.85%。预测结果表明，所建立的海底沉积物声速BP神经网络预测方法优于传统的单、二、多参数预测方程。