滑坡监测及辅助决策关键技术研究

发布时间：2018-03-02 09:42

本文关键词： 数据预处理机器学习滑坡区域检测局部区域滑坡预测滑坡辅助决策系统　出处：《江苏大学》2017年硕士论文　论文类型：学位论文

【摘要】：在我国,滑坡灾害是发生频率高、破坏性强、造成损失巨大的重大地质灾害之一。为了保障人民的生命财产安全,采取及时有效的手段进行滑坡监测预警是当前研究热点。现有的滑坡监测系统利用无线传感器网络技术,实现对滑坡的实时监测、数据实时传输及数据可视化,在人力无法完成的监测工作中发挥着重大作用。但系统缺乏对滑坡的区域性检测、整体性分析的功能和高效具有针对性的局部区域滑坡预测模型。因此,滑坡预测不及时,预测准确率不高。本文应用机器学习技术辅助系统对滑坡规律进行分析、识别、统计和滑坡等级划分,可间接辅助专家进行滑坡预警分析,提高滑坡预测准确率。主要完成的工作如下:1.采用数据预处理的去噪、平滑、归一化等方法对滑坡监测数据进行处理。对滑坡监测数据集进行数据清洗,剔除滑坡监测数据集中含有噪声、不一致、不完整、冗余和缺失的数据,进一步对数据集进行去噪、平滑和归一化处理,提高数据集质量,得到理想的训练数据集。2.结合了机器学习、R-tree、数据显著相关性等技术设计了一种基于机器学习技术的滑坡区域性检测方法。主要实现对预处理后的滑坡数据显著相关性评估,统计和计算各分片区域滑坡发生的概率,按照滑坡等级对各分片区域进行划分,自动标记并显示各分片区域滑坡概率数值。从而实现滑坡区域性检测、整体性分析的功能。3.提出了一种局部区域滑坡预测模型。该模型主要是使用改进的极限学习机(ELM)算法进行模型训练学习,并融合受限玻尔磁曼机对已建立的模型参数进行微调优化模型;借鉴了深度学习中卷积神经网络等技术的多隐层思想,在模型的训练过程中加入了多隐层的层数选择设置,不断将低层的数据特征传到上一层,直到最后一层,得到一个数据特征集。然后对整个数据特征集再反复训练学习,建立局部区域滑坡预测模型。经过对江苏省镇江市南山风景区、江心洲、百盛家园三个实际滑坡场景中运行的实验结果分析,可提高滑坡预测准确率。
[Abstract]:In China, landslide disaster is one of the major geological disasters with high frequency, strong destruction and huge losses. In order to ensure the safety of people's lives and property, The current landslide monitoring system uses wireless sensor network technology to realize real-time landslide monitoring, real-time data transmission and data visualization. It plays an important role in the monitoring work which can not be completed by manpower. However, the system lacks the regional detection of landslide, the function of holistic analysis and the efficient and targeted local landslide prediction model. Therefore, landslide prediction is not timely. The prediction accuracy is not high. This paper applies machine learning technology assistant system to analyze, identify, statistics and grade landslide analysis, which can indirectly assist experts in landslide early warning analysis. To improve the accuracy of landslide prediction. The main work accomplished is as follows: 1. The landslide monitoring data is processed by the methods of de-noising, smoothing and normalization of data preprocessing, and the data set of landslide monitoring is cleaned. The landslide monitoring data set contains noisy, inconsistent, incomplete, redundant and missing data. The data set is further de-noised, smoothed and normalized to improve the quality of the data set. An ideal training data set. 2. A regional landslide detection method based on machine learning technology is designed by combining machine learning R-tree and significant correlation of data. The evaluation of significant correlation of pre-processed landslide data is realized. Statistics and calculation of the probability of landslide occurrence in each segmented area are made. According to the grade of landslide, the probability values of landslide are automatically marked and displayed, so as to realize the regional landslide detection. The function of integrality analysis. 3. A local landslide prediction model is proposed. The model mainly uses the improved extreme learning machine (ELM) algorithm for model training and learning. The model is fine-tuned by the constrained Bohr Magneto Machine, and the multi-hidden layer idea of convolution neural network is used for reference, and the layer selection and setting of the multi-hidden layer are added in the training process of the model. The low-level data features are transmitted to the upper layer until the last layer, and a data feature set is obtained. Then the whole data feature set is trained and studied again and again. Based on the analysis of the experimental results in Nanshan Scenic spot Jiangxinzhou and Baisheng Jiayuan of Zhenjiang City Jiangsu Province the accuracy of landslide prediction can be improved.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P642.22;TP274;TP181

【相似文献】