基于稀疏自编码的集成学习台风灾害经济损失评估  

Economic loss assessment of typhoon disaster by integrated learning based on sparse self-coding

作　　者：徐新卫[1] 邓佳佳 陶飞 朱俊杰 周俊 XU Xinwei;DENG Jiajia;TAO Fei;ZHU Junjie;ZHOU Jun(College of Management Science and Engineering,Anhui University of Technology,Maanshan 243032,China;School of Management,Guangdong University of Science and Technology,Dongguan 523076,China)

机构地区：[1]安徽工业大学管理科学与工程学院,安徽马鞍山243032 [2]广东科技学院管理学院,广东东莞523076

出　　处：《海洋湖沼通报（中英文）》2024年第4期100-106,共7页Transactions of Oceanology and Limnology

基　　金：安徽省教育厅教学研究项目(2019jyxm0145)。

摘　　要：为了消除台风灾害数据指标间的多重共线性影响以及训练过程中的过拟合,本文使用稀疏自编码网络对广东省台风灾害数据进行非线性特征提取,作为BP神经网络的输入,最后通过Bagging方法将神经网络输出结果集成,以评估直接经济损失。为了验证稀疏自编码算法在特征提取方面的优势与有效性,本文引入主成分模型进行对比分析。试验结果表明稀疏自编码网络能够较好地提取非线性特征并进行评估,其评估精度优于以主成分分析为代表的线性降维方法,同时也证明了本文提出的集成神经网络的评估效果优于多层的BP神经网络,证明了此方法是一种解决台风灾害经济损失评估问题的有效方法,具有一定的应用参考价值。In order to eliminate the multi-collinearity between typhoon disaster data indicators and the over-fitting in the training process,this paper used sparse self-coding network to extract nonlinear features of typhoon disaster data in Guangdong province as the input of BP neural network.The output results of neural network were integrated by Bagging method to predict direct economic losses.In order to verify the advantage and effectiveness of sparse self-coding algorithm in feature extraction,the principal component model was introduced for comparative analysis.The experimental results showed that the sparse self-coding network could extract nonlinear features and predict them,and its prediction accuracy is better than that of the linear dimension reduction method represented by principal component analysis.It also proved that the prediction effect of the integrated neural network proposed in this paper was better than that of the multi-layer BP neural network.It was proved that this method was an effective method to solve the problem of economic loss assessment of typhoon disasters,and had a certain application reference value.

关 键 词：台风灾害损失 稀疏自编码器 BAGGING算法 BP神经网络 

分 类 号：P732[天文地球—海洋科学] X43[环境科学与工程—灾害防治]