ANN Based Relationship Modeling of Submarine Sedimentary Soil Properties and Resistivity

Resistivity measurement is one of the main methods for long-term in-situ observation of seabed engineering geology. It is necessary to establish the relationship between the submarine sedimentary physical parameters and the resistivity. The least square regression method is often used in traditional relationship modeling. Machine learning based data-driven method is a research hotspot in relationship modeling. In order to further improve the modeling accuracy, the relationship modeling between physical properties and resistivity of submarine sediments based on artificial neural network (ANN) is studied in this paper. According to the requirement of relationship modeling, the network structures of back propagation (BP) neural network, genetic algorithm optimization based BP neural network (GA-BP) and radial basis function (RBF) neural network are designed respectively. The experimental data of the northern slope of South China Sea obtained by “Experiment No. 3” scientific research vessel are used as training data samples, which are divided into training set, validation set and test set. Root mean square error (RMSE) and determination coefficient are used as evaluation indexes. The Matlab based simulation results show that the ANN based data-driven method can effectively establish the relationship model between physical properties and resistivity of seabed sediments, and the modeling accuracy is higher than that of the traditional least square regression method. RBF neural network has a better modeling effect.