基于门循环单元神经网络模型的煤层底板突水动态预测  

Dynamic Prediction of Water Inrush from Seam Floor Based on Gate Recurrent Unit Neural Network Model

作　　者：邓强 张召千[1] 王震 DENG Qiang;ZHANG Zhaoqian;WANG Zhen(College of Mining Engineering, Taiyuan University of Technology, Taiyuan 030024, China;Qinhuangdao Engineering Design and Research Institute Co., Ltd., Qinhuangdao 066000, China)

机构地区：[1]太原理工大学矿业工程学院,太原030024 [2]秦皇岛工程设计研究院有限公司,河北秦皇岛066000

出　　处：《太原理工大学学报》2021年第5期810-816,共7页Journal of Taiyuan University of Technology

摘　　要：在煤层底板突水理论及现场实测数据分析基础上,建立了煤层底板突水影响因素突水指标,并通过Wrapper评价策略的特征选择,筛选出了影响煤矿底板突水的主控因素。在对动态的煤层底板突水门循环单元神经网络模型进行训练并完成构建之后,将其与三种静态神经网络预测模型进行比较。结果表明:煤层底板突水门循环神经网络模型预测的准确率在训练、验证及测试阶段都高于静态神经网络预测模型,能够很好地完成煤层底板突水预测,提高煤矿生产安全。On the basis of the water inrush theory of coal mines and analysis of field measured data,this paper established water influx indicators for water inrush from coal seam floor.Through the feature selection of the Wrapper evaluation strategy,the main control factors affecting the water inrush of coal mine were finally selected.After training and establishing a dynamic gated recurrent unit(GRU)neural network prediction model of water inrush in coal seam,this model was compared with other static prediction models.The accuracy predicted by gate recurrent unit neural network model during the validation,training,and test phases was higher than that obtained with other static models,indicating that gate recurrent unit model can well predict the water inrush from coal seam floor and improve coal mine production safety.

关 键 词：煤层底板突水 特征选择 门循环单元神经网络 动态预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]