大数据环境下的滑坡危险性评估模型构建方法研究

The research of landslide hazard assessment model is the bottleneck problem of geological disaster risk control. It is expected that the big data technologies can be applied successfully to address this issue. Through the use of those technologies, this project aims to deal with some key problems, including landslide hazard indicator identification, landslide hazard assessment. Firstly, in order to identify landslide hazard indicator, the association analysis mining method is proposed on the basis of landslide spatial-temporal distribution regularity and hazard indicator relation, while coping with some key technologies in the relation analysis and intelligent extraction of multi-source multi-type heterogeneous landslide spatial-temporal big data. Then, by extracting elements automatically with relation analysis approach, the index system of large-area and generic-type landslide hazard assessment is also developed. Furthermore, in order to achieve dynamic modeling for landslide hazard assessment, the deep neural network is designed in accordance with adaptive deep learning algorithm. Specifically, in the proposed deep neural network, the complex relation between geological disaster environment and disaster hazard assessment index is extracted, and fast learning algorithm is also presented to deal with historically geological disaster big data. Moreover, under the Hadoop and Spark architecture, the parallel distributed computing framework is designed to implement model optimization and data parallelization. It is expected that research results of this project could play an important role in rapid early warning and monitoring of landslides.

滑坡危险性评估模型研究是地质灾害风险控制的瓶颈问题，大数据为此提供了研究契机。本项目围绕滑坡致灾因子识别、滑坡危险性评估等问题展开。首先针对滑坡致灾因子识别问题，提出基于滑坡时空分布规律及致灾因子关联关系分析挖掘技术，攻克基于多源、多类型异构滑坡时空大数据的关联关系分析及智能提取等关键技术。针对大区域泛类型滑坡危险性评估指标体系构建问题，基于关系分析法自动完成要素的抽取，构建滑坡危险性评估指标体系。针对滑坡危险性评估模型动态学习建立问题，研究自适应深度学习的深层神经网络，建立地质灾害环境到灾害危险性评估指标的映射关系，提出适应历史地质灾害大数据的快速学习算法，设计并行分布式计算框架（Hadoop、Spark），完成模型构建及数据并行化处理。项目预期成果对滑坡的快速预警监测具有重要研究价值。

本项研究针对目前大数据环境下滑坡危险性评估因子识别方法不足，动态非线性评估模型缺乏等突出问题，开展了大数据环境下的滑坡危险性评估因子自动识别和深度学习模型构建方法研究，研发了基于地质文本和遥感大数据的灾害信息智能提取、滑坡地质灾害时空分布规律及致灾因子自动关联分析挖掘等创新技术，对比分析并构建了一系列滑坡危险性评估深度学习模型，并基于研发的模型库管理与滑坡危险性评估等原型系统，围绕三类典型滑坡开展了示范验证，取得了良好的试点应用效果。总体研究成果构建形成了基于大数据的滑坡危险性评估技术方法体系，为滑坡灾害危险性评估提供了数据驱动下的新模式。