基于多目标优化方法的滑坡易发性评价

目的] 在滑坡易发性评价中,滑坡预测模型的选取和优化对运算过程的高效性和预测结果的准确性至关重要。针对现有单目标遗传优化算法(genetic algorithm,GA)易陷入早熟、局部搜索能力差、全局优化速度慢等问题,拟提出一种新的优化算法框架,将多目标遗传算法中的经典算法—带精英选择策略的非支配排序算法(the nondominated sorting genetic algorithm with an elite strategy,NSGA-Ⅱ)与常用机器学习模型随机森林(random forest,RF)、支持向量机(support vector machine,SVM)]相结合,进行滑坡易发性预测。与单目标优化不同的是,NSGA-Ⅱ算法可同时进行特征选择和超参数优化,并使预测模型同时实现最优准确度、召回率、精密度和AUC(area under curve,AUC)。方法] 以三峡库区重庆段为研究区,从模型精度评价、滑坡灾害易发性分区图、分区统计3个方面对4种优化模型(RF-GA、SVM-GA、RF-NSGA-II、SVM-NSGA-II)进行对比分析。结果] NSGA-II较GA优化效果更明显,在模型评价和滑坡易发性分区方面,RF-NSGA-II模型具有更高的预测性能,4项评价值分别为80.91%,81.89%,80.07%,88.60%,证明NSGA-II优化算法的有效性;极低至极高危险区面积占比依次为23.06%,22.46%,22.96%,19.99%,11.53%,验证了RF-NSGA-II模型的可靠性。由RF-NSGA-II模型预测得到的易发性图表明,高和极高易发性区集中在研究区北部,且由东向西呈带状分布。结论] 研究采取的基于多目标选择的RF-NSGA-II模型,为滑坡易发性评价中机器学习模型调优提供新思路。

Evaluation of Landslide Susceptibility Based on Multi-objective Optimization Method

Objective] In the landslide susceptibility assessment, the selection and optimization of the landslide prediction model are very important to the efficiency of the calculation process and the accuracy of the prediction results. Aiming at the problems that the existing single-objective genetic optimization algorithm (Genetic Algorithm, GA) is prone to premature maturity, poor local search ability, and slow global optimization speed, this paper develops a new optimization algorithm framework, which integrates the classic algorithm in the multi-objective genetic algorithm-Non-dominated sorting method with elite selection strategy (NSGA-II) combined with common machine learning algorithms, random forest (RF), and support vector machine (SVM) to predict landslide susceptibility. Different from single objective optimization, NSGA-II algorithm can perform feature selection and hyper-parameter optimization simultaneously, and make the prediction model achieve optimal accuracy, recall, precision and AUC (area under curve, AUC) at the same time. Methods] Taking the Chongqing section of the Three Gorges reservoir area as the study area, the four optimized models (RF-GA, SVM-GA, RF-NSGA-II. and SVM-NSGA-II) were compared and analyzed in three aspects: model accuracy evaluation, landslide hazard susceptibility zoning map, and zoning statistics. Results] NSGA-II was more effective than GA optimization, and in terms of model evaluation and landslide susceptibility zoning, the RF-NSGA-II model had higher predictive performance, with four evaluation values of 80.91%, 81.89%, 80.07% and 88.60% respectively, proving the effectiveness of the NSGA-II optimization algorithm; the area share of very low to very high hazard zones were in the order of 23.06%, 22.46%, 22.96%, 19.99%, and 11.53%, which verified the reliability of the RF-NSGA-II model. The susceptibility map predicted by the RF-NSGA-II model showed that the high and extremely high susceptibility areas were concentrated in the north and distributed in bands from east to west. Conclusion] RF-NSGA-II algorithm based on multi-objective selection provides a new idea for the optimization of machine learning model for landslide risk assessment.