中国安全科学学报 ›› 2019, Vol. 29 ›› Issue (5): 37-43.doi: 10.16265/j.cnki.issn1003-3033.2019.05.007

• 安全系统学 • 上一篇    下一篇

基于QRNN模型的生命年损失概率密度预测

武佳佳1, 王威**2 副研究员, 朱强强1, 马东辉2 研究员   

  1. 1 北京工业大学 建筑工程学院,北京 100124;
    2 北京工业大学 建筑与城市规划学院,北京 100124
  • 收稿日期:2019-02-02 修回日期:2019-04-02 发布日期:2020-11-02
  • 通讯作者: **王威(1981—),男,河南沈丘人,博士,副研究员,主要从事城市安全与防灾规划、生命线系统抗灾评价技术等方面的研究。E-mail:ieeww@bjut.edu.cn。
  • 作者简介:武佳佳(1992—),女,河北邯郸人,博士研究生,主要研究方向为生命线工程抗震防灾。E-mail:snail_s@emails.bjut.edu.cn。
  • 基金资助:
    国家自然科学基金资助(51678017);国家重点研发计划课题(2018YFD1100902-1)。

Lifeyears loss probability density prediction based on QRNN model

WU Jiajia1, WANG Wei2, ZHU Qiangqiang1, MA Donghui2   

  1. 1 College of Architecture and Civil Engineering, Beijing University of Technology, Beijing 100124, China;
    2 College of Architecture and Urban Planning, Beijing University of Technology, Beijing 100124, China
  • Received:2019-02-02 Revised:2019-04-02 Published:2020-11-02

摘要: 为全面评估预测震害损失,提出一种概率密度预测方法。首先,通过改进的生命年损失计算法,获取生命年损失值;其次,采用基于Akaike信息量准则(AIC)的逐步回归分析法,辨识生命年损失的强关联因素,在此基础上构建神经网络分位数回归(QRNN)模型;然后,得到生命年损失预测值与强相关因素的非线性关系,输出不同分位点下生命年损失预测值,运用高斯核函数预测生命年损失概率密度;最后,选取我国1996—2014年的189条地震灾害损失数据作为训练样本,预测2015年10例地震的生命年损失,并与B样条分位数回归(QRBS)模型及3种线性模型作对比。研究表明:基于QRNN模型的震害损失评估概率密度预测,降低了数据依赖性,提高了评估效率;预测值平均绝对误差不超过7.5%,便于震害评估。

关键词: 生命年损失, 逐步回归, 神经网络分位数回归(QRNN), 关联因素辨识, 概率密度函数

Abstract: In order to comprehensively assess and predict earthquake damage losses, a method for probability density prediction is proposed. First, the life-year loss was obtained through the improved lifeyears loss calculation method. Secondly, by use of stepwise regression analysis which is based on Akaike information criterion (AIC), the strongly correlated factors of lifeyears loss were identified and furthermore QRNN model was constructed. Then the nonlinear relationship between the predicted value of lifeyears loss and the strongly correlated factors was obtained, the predicted loss under different quantile points was outputted, and the lifeyears loss probability density was predicted by adopting Gaussian kernel function. Finally, with the damage loss data of 189 Chinese earthquakes from 1996 to 2014 as training samples, the lifeyears loss of 10 earthquakes in 2015 was predicted and compared with quantile regression B-spline (QRBS) model and three linear models. The results show that the damage loss probability density prediction based on the proposed model reduces data dependency while improves evaluation efficiency, and the average absolute error of the prediction is less than 7.5%, which is effective for damage assessment.

Key words: lifeyears loss, stepwise regression, quantile regression neural network (QRNN), correlated factors identification, probability density function

中图分类号: