基于优化长短期记忆网络的矿坑遗产沉降预测-《计算机技术与发展》

文章信息/Info

Title:: Settlement Prediction for Mine Heritage Based on Optimized Long Short Term Memory Network

作者:: 王凤英1; 2; 孟令泽1; 哈静1; 杜利明1; 2; 1. 沈阳建筑大学计算机科学与工程学院,辽宁沈阳 110168; 2. 宿迁学院信息工程学院,江苏宿迁 223800

Author(s):: WANG Feng-ying1; 2; MENG Ling-ze1; HA Jing1; DU Li-ming1; 2; 1. School of Computer Scienceand Engineering,Shenyang Jianzhu University,Shenyang 110168,China; 2. School of Information Engineering,Suqian University,Suqian 223800,China

Keywords:: industrial mine heritage; settlement prediction; warning model; long short-term memory network; Dung beetle optimization

摘要:: 工业矿坑遗产以其独特风貌和价值逐步受到广泛关注。针对矿坑遗产易发的沉降地质灾害,积极采取预防措施是降低损失的有效途径。为解决工业矿坑遗产沉降灾害预测问题,提出一种融合蜣螂优化算法(DBO)的优化长短期记忆网络(LSTM)算法,用于构建预警模型。选取阜新市海州露天矿作为实验地点,利用小基线集合成孔径雷达干涉测量(SBAS-InSAR)技术采集 55 景矿区沉降数据。通过两种去噪方法对采集到的样本数据进行去噪处理,应用 DBO 算法优化 LSTM,建立工业矿坑遗产沉降预测模型。 LSTM 模型的超参数使用 DBO 算法优化以实现高精度预测模型,并与其他算法优化 LSTM 后的模型指标进行对比。结果表明:DBO-LSTM 模型在工业矿坑遗产沉降预测优势突出,预测模型的均方根误差、平均绝对误差和决定系数分别为 0. 045 mm,0. 038 mm,0. 956,均优于其他预测模型。 DBO-LSTM 模型在预测工业矿坑遗产沉降方面展现了高精度、快速收敛和强稳定性等特点,为工业矿坑遗产保护工作提供了有力支持。

Abstract:: Industrial mine heritage has gradually gained attention for its unique value. In response to the geological hazards of subsidence in mining heritage sites,taking proactive preventive measures is an effective way to reduce losses. We propose a Long Short -Term Memory network (LSTM) integrated with Dung Beetle Optimizer (DBO) to construct a settlement warning model for industrial mine heritage. Selecting the Haizhou open-pit mine in Fuxin City as the experimental site,the small baseline set synthetic aperture radar inter-ferometry (SBAS-InSAR) technology was used to collect settlement data from 55 mining areas. Two denoising methods were used to denoise the collected sample data,and the DBO was applied to optimize the LSTM and establish an industrial mine heritage settlement prediction model. The hyperparameters of the LSTM model were optimized using the DBO to achieve high-precision prediction models,and compared with the model metrics optimized by other algorithms. The experimental results show that the new model has outstanding advantages in predicting the settlement of industrial mine heritage,with low root mean square error 0. 045 mm,mean absolute error 0. 038mm and a determination coefficient of 0. 956 respectively. It demonstrates high precision, fast convergence and strong stability in predicting the settlement of industrial mine heritage,which provides strong support for the protection of industrial mine heritage.