为快速评估地震直接经济损失,针对我国西部地区,尝试采用随机森林机器学习回归算法,以1993—2017年震害数据为基础,结合各年份经济数据与抗震设计数据,经特征选择与参数优化后,进行模型的训练与测试。实验结果表明,在减少模型输入特征的情况下,优化后的随机森林模型可得到更优的评估结果。通过删除含有缺失特征样本的数据预处理方法,评估模型的决定系数R²达到0.86,优于中值补齐缺失特征数据预处理下的评估模型,更适用于地震直接经济损失的评估。实例验证表明该模型评估结果与实际经济损失有较好的一致性,可为抗震救灾提供决策支持。

In this study,we aim to expedite the assessment of direct economic losses induced by earthquakes,focusing on China's western region. We employ a random forest machine learning regression algorithm for this purpose. Leveraging earthquake damage data spanning from 1993 to 2017,in conjunction with economic and seismic design data from various years,we train and test the model following feature selection and parameter optimization steps. The findings reveal that the optimized random forest model yields superior evaluation outcomes while reducing the model's input features. Specifically,the evaluation model achieves an R² value of 0.86 under the data preprocessing method involving the deletion of missing feature samples,surpassing the evaluation model's performance under the median filling missing feature data preprocessing approach. This optimized model proves more suitable for assessing direct economic losses attributable to earthquakes. Validation using real world examples demonstrates that the evaluation results derived from this model align closely with actual economic losses,underscoring its utility in providing decision support for earthquake relief efforts.

P315

国家重点研发计划(2018YFC1504506)资助