~采空区地面塌陷是煤矿企业常见的地质灾害之一，它破坏范围广、影响大、持续时间长，为了降低采空区冒落、失稳造成的重大经济损失和人员伤亡，研究煤矿采空区地面塌陷危险性评价，对煤矿企业安全生产尤为重要。本文将深度学习应用到煤矿采空区地面塌陷危险性评价中，并将其结果进行三维可视化。具体研究工作如下：
(1) 首先分析了采空区地面塌陷发生机理，通过对煤矿的地质勘探钻孔数据、水文数据等进行提取、分析，用分位图法进行正态性检验，得出了影响采空区稳定性的主要因素。为了使用深度学习模型，采用基于封装式评价方法进行了特征选择、利用BP神经网络处理采空区顶板岩性数据的缺失值、选择归一化方法对数据进行预处理。
(2) 针对传统采空区危险性评价方法不能处理时间序列数据的问题，本文使用深度学习中的长短时记忆网络（Long Short-Term Memory, LSTM）作为采空区地面塌陷危险性评价模型。首先将预处理后的影响采空区塌陷的10个特征作为LSTM模型输入，根据输入的数据量，经过实验选取每批处理的样本个数为8，通过对SGD、RMSprop和Adam的比对选取Adam作为优化函数，隐藏层节点数设定为30，Dropout值为0.5时LSTM模型能够达到最佳性能。将模型的输出分为采空区危险性评价的四个等级，从而构建基于LSTM的采空区地面塌陷危险性评价模型。为了验证模型在解决这类问题上的效果和性能，将得到的采空区危险性评价结果与BP神经网络模型和SVM支持向量机的评价结果进行了实验对比，结果表明，本模型的评价准确度比BP网络模型提高5.44%，比常规SVM模型提高4.1%，进一步说明了本模型在煤矿采空区地面塌陷危险性评价中的优势。
(3) 为了直观地展示采空区危险性评价结果，利用GOCAD软件构建地质数据库采用建立点、线、面的方法构建地质三维模型，使用Python语言和本文模型对陕北某煤矿采空区地面塌陷危险性进行评价，并将预测结果三维可视化。
 

~Ground subsidence in the goaf is one of the common geological disasters in coal mining enterprises. It has a wide range of damage, great impact and long duration. In order to reduce the major economic losses and casualties caused by the fall of the goaf and instability, the study of coal mine goaf The evaluation of the risk of ground collapse in the district is particularly important for the safe production of coal mine enterprises. In this paper, deep learning is applied to the evaluation of the risk of ground collapse in the goaf area of coal mines, and the results are visualized in three dimensions. The specific research work is as follows:
(1) Firstly, the mechanism of ground subsidence in the mined-out area is analyzed. Through the extraction and analysis of the geological exploration drilling data and hydrological data of the coal mine, the normality test is carried out with the quantile map method, and the influence of the mined-out area is obtained. The main factor of stability. In order to use the deep learning model, feature selection based on encapsulated evaluation strategy, BP neural network to deal with missing values, and normalization method are selected to preprocess the data.
(2) Aiming at the problem that traditional goaf risk assessment methods cannot handle time series data, the Long Short-Term Memory (LSTM) in deep learning is used as the risk evaluation model of the goaf ground collapse. First, the 10 features that affect the collapse of the goaf after preprocessing are input as the LSTM model. According to the amount of input data, the number of samples for each batch is selected as 8 through experiments, and selected by comparing SGD, RMSprop and Adam. Adam is used as an optimization function, the number of hidden layer nodes is set to 30, and the dropout value is 0.5 when the LSTM model can achieve the best performance. The output of the model is divided into four levels of the risk evaluation of the mined-out area, so as to construct the risk evaluation model of the mined-out area ground collapse based on LSTM. In order to verify the excellent performance of the model in solving such problems, the obtained goaf risk evaluation results were compared with the evaluation results of the BP neural network model and SVM support vector machine. The results show that the evaluation accuracy of the model is 5.44% higher than the BP network model, and 4.1% higher than the conventional SVM model, which further illustrates the accuracy of this model in the evaluation of the risk of ground collapse in coal mined areas.
(3) In order to visually display the results of the risk assessment of the mined-out area, use GOCAD software to build a geological database, use the method of establishing points, lines, and surfaces to build a three-dimensional geological model. The collapse risk is evaluated, and the prediction results are visualized in a three-dimensional map.