基于深度神经网络的滑坡危险性评价——以深圳市为例
|
摘要
该文以提升滑坡危险性评价精度为核心目标,对深度神经网络在滑坡危险性评价中的可行性和适用性进行研究,以期充分发挥深度神经网络强大的非线性学习和拟合能力,取得更加合理的滑坡危险性评价结果。选取滑坡灾害多发的深圳市作为实例,基于深圳市815条历史滑坡数据,开展了深度神经网络建模训练;通过与广义线性模型及分类与回归树模型训练效果的对比,对深度神经网络的建模效果进行了评价,深度神经网络、广义线性模型和分类与回归树模型的AUC值依次是0.908、0.861和0.857。将训练所得的模型应用于深圳市全区,对3种模型输出的滑坡危险性评价成果的合理性和可靠性进行了对比分析,结果表明:深度神经网络建模精度良好,优于常见的广义线性模型和分类与回归树模型,输出的滑坡危险性评价成果具有合理性,适用于滑坡危险性评价工作。
Landslide susceptibility assessment is considered to be a key role in landslide risk management tasks.In order to improve the rationality and accuracy of the landslide susceptibility assessment,deep neural network(DNN) is proposed to be applied in landslide susceptibility assessment to take advantage of its excellent nonlinear learning and generalization ability in this paper,which has been proven to be necessary and crucial in landslide susceptibility model training.Taking Shenzhen City,where landslide hazards occurred frequently,as the study area,a DNN model with 13 layers was designed and trained based on 815 historical landslide occurrence records in the study area.The performance of deep neural network was then analyzed and compared with two typical models,generalized linear model(GLM) and classification and regression trees(CART).The AUC of deep neural network reaches 0.908 while GLM and CART get 0.861 and 0.857 respectively,which means that DNN model outperforms GLM or CART in this case study.Then,all the above three models were applied to the whole study area to carry out further evaluation and comparison of the reasonability and reliability of the outputs by the three models,both in general area and in three typical landform regions.It can be concluded that DNN achieves better accuracy than GLM and CART.The resulting map of DNN is considerately reasonable and credible,which shows deep neural network is overall suitable for landslide susceptibility assessment.
Landslide susceptibility assessment is considered to be a key role in landslide risk management tasks.In order to improve the rationality and accuracy of the landslide susceptibility assessment,deep neural network(DNN) is proposed to be applied in landslide susceptibility assessment to take advantage of its excellent nonlinear learning and generalization ability in this paper,which has been proven to be necessary and crucial in landslide susceptibility model training.Taking Shenzhen City,where landslide hazards occurred frequently,as the study area,a DNN model with 13 layers was designed and trained based on 815 historical landslide occurrence records in the study area.The performance of deep neural network was then analyzed and compared with two typical models,generalized linear model(GLM) and classification and regression trees(CART).The AUC of deep neural network reaches 0.908 while GLM and CART get 0.861 and 0.857 respectively,which means that DNN model outperforms GLM or CART in this case study.Then,all the above three models were applied to the whole study area to carry out further evaluation and comparison of the reasonability and reliability of the outputs by the three models,both in general area and in three typical landform regions.It can be concluded that DNN achieves better accuracy than GLM and CART.The resulting map of DNN is considerately reasonable and credible,which shows deep neural network is overall suitable for landslide susceptibility assessment.
引文
[1] 殷坤龙,朱良峰.滑坡灾害空间区划及GIS应用研究[J].地学前缘,2001,8(2):279-284.
[2] 田原.基于多因子统计模型的滑坡危险性区划方法研究[D].北京:北京大学,2010.
[3] ATKINSON P M,MASSARI R.Generalised linear modelling of susceptibility to landsliding in the Central Apennines,Italy[J].Computers & Geosciences,1998,24(4):373-385.
[4] SAITO H,NAKAYAMA D,MATSUYAMA H.Comparison of landslide susceptibility based on a decision-tree model and actual landslide occurrence:The Akaishi Mountains,Japan[J].Geomorphology,2009,109(3):108-121.
[5] 朱良峰,吴信才,殷坤龙,等.基于信息量模型的中国滑坡灾害风险区划研究[J].地球科学与环境学报,2004,26(3):52-56.
[6] 马志江,陈汉林,杨树锋.基于支持向量机理论的滑坡灾害预测——以浙江庆元地区为例[J].浙江大学学报(理学版),2003,30(5):592-596.
[7] BRENNING A.Statistical geocomputing combining R and SAGA:The example of landslide susceptibility analysis with generalized additive models[J].SAGA-Seconds Out,2008,19(23-32):410.
[8] LEE S,RYU J H,MIN K,et al.Landslide susceptibility analysis using GIS and artificial neural network[J].Earth Surface Processes and Landforms,2003,28(12):1361-1376.
[9] 司康平.基于GIS的滑坡灾害空间预测:方法比较及因子重要性分析——以深圳市为例[D].北京:北京大学,2009.
[10] 司康平,田原,汪大明,等.滑坡灾害危险性评价的3种统计方法比较——以深圳市为例[J].北京大学学报(自然科学版),2008,54(4):639-646.
[11] LECUN Y,BENGIO Y,HINTON G.Deep learning[J].Nature,2015,521(7553):436-444.
[12] LECUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[13] XIA F,LIU T Y,WANG J,et al.Listwise approach to learning to rank:Theory and algorithm[A].Proceedings of the 25th International Conference on Machine learning[C].2008.1192-1199.
[14] BOTTOU L E,ON,LECUN Y.Large scale online learning[J].Advances in Neural Information Processing Systems,2004,16:217.
[15] KINGMA D,BA J.Adam:A method for stochastic optimization[A].Proceedings of the International Conference on Learning Representations[C].2015.1-15.
[16] 李亭,田原,邬伦,等.基于随机森林方法的滑坡灾害危险性区划[J].地理与地理信息科学,2014,30(6):25-30.
[17] 熊浪涛,田原,刘鹏.滑坡危险性区划中基于一类分类模型的样本筛选[J].地理与地理信息科学,2016,32(3):43-48.
[18] MARQUARIDT D W.Generalized inverses,ridge regression,biased linear estimation,and nonlinear estimation[J].Technometrics,1970,12(3):591-612.
[19] IOFFE S,SZEGEDY C.Batch normalization:Accelerating deep network training by reducing internal covariate shift[A].Proceedings of the 32nd International Conference on Machine Learning[C].2015.448-456.
[20] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenet classification with deep convolutional neural networks[A].Advances in Neural Information Processing Systems Conference[C].2012.1097-1105.
[21] GLOROT X,BORDES A,BENGIO Y.Deep sparse rectifier neural networks[A].Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics[C].2011.315-323.
[22] 邢秋菊,赵纯勇,高克昌,等.基于GIS的滑坡危险性逻辑回归评价研究[J].地理与地理信息科学,2004,20(3):49-51.
[23] HAN J,KAMBER M,PEI J.Data Mining:Concepts and Techniques[M].San Francisco,USA:Morgan Kaufmann Publishers Inc.,2011.
[24] MCLACHLAN G,DO K A,AMBROISE C.Analyzing Microarray Gene Expression Data[M].New York,USA:John Wiley & Sons,2005.
[25] TOBLER W R.A computer movie simulating urban growth in the Detroit region[J].Economic Geography,1970,46(sup1):234-240.
[2] 田原.基于多因子统计模型的滑坡危险性区划方法研究[D].北京:北京大学,2010.
[3] ATKINSON P M,MASSARI R.Generalised linear modelling of susceptibility to landsliding in the Central Apennines,Italy[J].Computers & Geosciences,1998,24(4):373-385.
[4] SAITO H,NAKAYAMA D,MATSUYAMA H.Comparison of landslide susceptibility based on a decision-tree model and actual landslide occurrence:The Akaishi Mountains,Japan[J].Geomorphology,2009,109(3):108-121.
[5] 朱良峰,吴信才,殷坤龙,等.基于信息量模型的中国滑坡灾害风险区划研究[J].地球科学与环境学报,2004,26(3):52-56.
[6] 马志江,陈汉林,杨树锋.基于支持向量机理论的滑坡灾害预测——以浙江庆元地区为例[J].浙江大学学报(理学版),2003,30(5):592-596.
[7] BRENNING A.Statistical geocomputing combining R and SAGA:The example of landslide susceptibility analysis with generalized additive models[J].SAGA-Seconds Out,2008,19(23-32):410.
[8] LEE S,RYU J H,MIN K,et al.Landslide susceptibility analysis using GIS and artificial neural network[J].Earth Surface Processes and Landforms,2003,28(12):1361-1376.
[9] 司康平.基于GIS的滑坡灾害空间预测:方法比较及因子重要性分析——以深圳市为例[D].北京:北京大学,2009.
[10] 司康平,田原,汪大明,等.滑坡灾害危险性评价的3种统计方法比较——以深圳市为例[J].北京大学学报(自然科学版),2008,54(4):639-646.
[11] LECUN Y,BENGIO Y,HINTON G.Deep learning[J].Nature,2015,521(7553):436-444.
[12] LECUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[13] XIA F,LIU T Y,WANG J,et al.Listwise approach to learning to rank:Theory and algorithm[A].Proceedings of the 25th International Conference on Machine learning[C].2008.1192-1199.
[14] BOTTOU L E,ON,LECUN Y.Large scale online learning[J].Advances in Neural Information Processing Systems,2004,16:217.
[15] KINGMA D,BA J.Adam:A method for stochastic optimization[A].Proceedings of the International Conference on Learning Representations[C].2015.1-15.
[16] 李亭,田原,邬伦,等.基于随机森林方法的滑坡灾害危险性区划[J].地理与地理信息科学,2014,30(6):25-30.
[17] 熊浪涛,田原,刘鹏.滑坡危险性区划中基于一类分类模型的样本筛选[J].地理与地理信息科学,2016,32(3):43-48.
[18] MARQUARIDT D W.Generalized inverses,ridge regression,biased linear estimation,and nonlinear estimation[J].Technometrics,1970,12(3):591-612.
[19] IOFFE S,SZEGEDY C.Batch normalization:Accelerating deep network training by reducing internal covariate shift[A].Proceedings of the 32nd International Conference on Machine Learning[C].2015.448-456.
[20] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenet classification with deep convolutional neural networks[A].Advances in Neural Information Processing Systems Conference[C].2012.1097-1105.
[21] GLOROT X,BORDES A,BENGIO Y.Deep sparse rectifier neural networks[A].Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics[C].2011.315-323.
[22] 邢秋菊,赵纯勇,高克昌,等.基于GIS的滑坡危险性逻辑回归评价研究[J].地理与地理信息科学,2004,20(3):49-51.
[23] HAN J,KAMBER M,PEI J.Data Mining:Concepts and Techniques[M].San Francisco,USA:Morgan Kaufmann Publishers Inc.,2011.
[24] MCLACHLAN G,DO K A,AMBROISE C.Analyzing Microarray Gene Expression Data[M].New York,USA:John Wiley & Sons,2005.
[25] TOBLER W R.A computer movie simulating urban growth in the Detroit region[J].Economic Geography,1970,46(sup1):234-240.