基于谱聚类算法的海底滑坡危险性评价

海底滑坡的危险性评价与分区，对海洋工程设施的选址和危险预防具有指导作用。本文基于无监督机器学习的谱聚类算法对黄河口埕岛海域展开了海底滑坡危险性评价，构建了输入参数为9、输出类别为4、核函数参数为0.08的海底滑坡危险性评价模型。使用该模型进行评价，将研究区分为了海底滑坡危险性高、较高、较低和低的区域。评价结果与地质环境因素分布特征对比显示，最重要的影响因素为海底沉积物类型和水动力作用，最重要的触发因子为液化。模型参数分析结果显示，合理简化输入因子可获得精度略低的评价结果，而核函数参数是影响评价准确性的重要指标。以上研究表明，谱聚类算法能够较好地用于海底滑坡危险性评价，数据类别丰富度和精度是影响评价精细程度的重要因素。

Risk assessment of submarine landslide based on spectral clustering

The risk assessment and zoning of submarine landslides can guide the site selection and risk prevention of offshore engineering facilities.In this paper,an unsupervised machine learning spectral analysis algorithm was used to evaluate the risk of submarine landslides in the Chengdao sea area of the Yellow River Estuary.A model of submarine landslides risk assessment with 9 input parameters,4 output parameters and 0.08 as kernel function parameters is constructed.By using this model,the study area can be divided into 4 parts:high,quite high,quite low and low risk of submarine landslide.The comparison between the evaluation results and the distribution characteristics of geological environment factors show that the most important factors are the type of seafloor sediment and hydrodynamic action,and the most important trigger factor is liquefaction.The analysis results of model parameters present that the evaluation results with slightly lower accuracy can be obtained by reasonably simplifying the input factors,and the kernel function parameter is important index affecting the evaluation accuracy.The above research shows that the unsupervised machine learning algorithm can be well used in the risk assessment of submarine landslides,and the richness and accuracy of data categories are important factors affecting the assessment accuracy.