基于CNN-aGRU融合模型的尾矿坝浸润线预测方法

doi:10.16265/j.cnki.issn1003-3033.2023.S1.2481

中国安全科学学报 ›› 2023, Vol. 33 ›› Issue (S1): 119-127.doi: 10.16265/j.cnki.issn1003-3033.2023.S1.2481

基于CNN-aGRU融合模型的尾矿坝浸润线预测方法

阮顺领¹^,²(), 韩思淼¹^,²^,^**(), 张宁宁¹, 顾清华¹^,², 卢才武¹^,³

¹ 西安建筑科技大学资源工程学院, 陕西西安 710055
² 西安建筑科技大学矿山系统工程研究所, 陕西西安 710055
³ 西安市智慧工业感知计算与决策重点实验室, 陕西西安 710055

收稿日期:2023-02-14 修回日期:2023-05-08 出版日期:2023-06-30
通讯作者:
**韩思淼(1999—),女,安徽阜阳人,硕士研究生,研究方向为矿山安全智能分析。E-mail:Hansimiao68@163.com。
作者简介:
阮顺领 (1981—),男,河南周口人,博士,副教授,主要从事矿山系统优化与管理方面的研究。E-mail:ruanshunling@163.com。
顾清华教授
卢才武教授
基金资助:
国家自然科学基金面上项目(52374160); 陕西省自然科学基础研究计划项目(2022JM-201)

Prediction method of saturation line of tailings dam based on CNN-aGRU fusion model

RUAN Shunling¹^,²(), HAN Simiao¹^,²^,^**(), ZHANG Ningning¹, GU Qinghua¹^,², LU Caiwu¹^,³

¹ School of Resources Engineering, Xi'an University of Architecture and Technology, Xi'an Shaanxi 710055, China
² Institute of Mine Systems Engineering, Xi'an University of Architecture and Technology, Xi'an Shaanxi 710055, China
³ Xi'an Key Laboratory of Perceptive Computing and Decision for Intelligent Industry, Xi'an Shaanxi 710055, China

Received:2023-02-14 Revised:2023-05-08 Published:2023-06-30

摘要/Abstract

摘要：

为预防尾矿库溃坝安全事故,关联分析在线监测数据的深度,提出将卷积神经网络(CNN)与门控循环单元(GRU)融合的尾矿库浸润线安全态势预测方法,以掌握坝体的稳定性状况与安全发展态势。该方法综合考虑尾矿库监测数据复杂非线性和时序关联性等特点,利用一维卷积神经网络(1D CNN)模型获取多源数据的局部关联特征和空间特征,并利用GRU模型获取浸润线数据的时序特征,采用自适应矩估计权重衰减优化算法(AdamW)优化模型梯度的自适应性,提高预测模型泛化能力和预测精度,并以河南洛阳某金属露天钼矿尾矿坝进行试验验证。试验结果表明:对比传统BP神经网络、长短期记忆网络(LSTM)、GRU等预测模型,该预测模型在平均绝对百分比误差(MAPE)、平均绝对误差(MAE)、均方误差(MSE)、均方根误差(RMSE)、R²几项关键指标上,分别达到0.013 915 62、0.005 432、0.000 045、0.006 702、0.998 334,实现对浸润线变化态势快速精准预测。

关键词: 卷积神经网络(CNN), 门控循环单元(GRU), 尾矿坝, 浸润线, 自适应矩估计权重衰减优化算法(AdamW)

Abstract:

In order to prevent the safety accident caused by tailings dam breaks, correlation analysis was conducted on the depth of online monitoring data. A safety situation prediction method of the saturation line of the tailings dam was proposed, which integrated CNN and GRU, so as to grasp the stability and safety development trend of the dam. By comprehensively considering the complex nonlinear and time-series correlation characteristics of monitoring data of the tailings dam, a one-dimensional CNN network was introduced to obtain the local correlation characteristics and spatial characteristics among the multi-source data. Then the GRU model was used to extract the time-series characteristics of the saturation line data. AdamW was used to optimize the adaptability of the model gradient, so as to improve the generalization ability and prediction accuracy of the prediction model. The method was tested and verified in the tailings dam of a metal open-pit molybdenum mine in Luoyang, Henan Province. The results show that compared with the traditional back propagation (BP) neural network, long short-term memory (LSTM) network, GRU, and other prediction models, the proposed prediction model reaches 0.013 915 62, 0.005 432, 0.000 045, 0.006 702, and 0.998 334 for the key indices of mean absolute percentage error (MAPE), mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and R², respectively. The method thus can rapidly and accurately predict saturation line changes.

Key words: convolutional neural network (CNN), gated recurrent unit (GRU), tailings dam, saturation line, Adam weight decay optimizer (AdamW)

阮顺领, 韩思淼, 张宁宁, 顾清华, 卢才武. 基于CNN-aGRU融合模型的尾矿坝浸润线预测方法[J]. 中国安全科学学报, 2023, 33(S1): 119-127.

RUAN Shunling, HAN Simiao, ZHANG Ningning, GU Qinghua, LU Caiwu. Prediction method of saturation line of tailings dam based on CNN-aGRU fusion model[J]. China Safety Science Journal, 2023, 33(S1): 119-127.

图/表 12

图1

表1

表2

图2

图3

表3

表4

表5

表6

表7

表8

图4

参考文献 23

[1]	于广明, 宋传旺, 潘永战, 等. 尾矿坝安全研究的国外新进展及我国的现状和发展态势[J]. 岩石力学与工程学报, 2014, 33(增1):3 238-3 248.
	YU Guangming, SONG Chuanwang, PAN Yongzhan, et al. Review of new progress in tailing dam safety in foreign research and current state with development trend in China[J]. Chinese Journal of Rock Mechanics and Engineering, 2014, 33(S1): 3 238-3 248.
[2]	HUI Shiqiang, CHARLEBOIS L, SUN Colin. Real-time monitoring for structural health, public safety, and risk management of mine tailings dams[J]. Canadian Journal of Earth Sciences, 2018, 55(3): 221-229. doi: 10.1139/cjes-2017-0186
[3]	张家荣, 刘建林. 中国尾矿库溃坝与泄露事故统计及成因分析[J]. 中国钼业, 2019, 43(4): 10-14.
	ZHANG Jiarong, LIU Jianlin. The statistics and causes of dam break and leakage in Chinese tailings pond[J]. China Molybdenum Industry, 2019, 43(4):10-14.
[4]	曾纪涵, 章光, 吴浩, 等. 改进POT模型下尾矿坝综合预警值确定方法[J]. 中国安全科学学报, 2022, 32(5):134-139. doi: 10.16265/j.cnki.issn1003-3033.2022.05.1699
	ZENG Jihan, ZHANG Guang, WU Hao, et al. Determination method of comprehensive early-warning indexes for tailings dam based on improved POT model[J]. China Safety Science Journal, 2022, 32(5):134-139. doi: 10.16265/j.cnki.issn1003-3033.2022.05.1699
[5]	阳雨平, 黄丕森, 陈国国. 基于改进FIM-未确知测度的尾矿库风险评价模型及应用[J]. 安全与环境学报, 2021, 21(3):996-1 004.
	YANG Yuping, HUANG Pisen, CHEN Guoguo. Safety evaluation of tailings pond based on FIM-optimization unascertained measure[J]. Journal of Safety and Environment, 2021, 21(3):996-1 004.
[6]	张岳安, 周科平. 浸润线埋深对尾矿坝稳定性的指标满意度敏感性分析[J]. 中国安全科学学报, 2013, 23(10):93-99.
	ZHANG Yuean, ZHOU Keping. Sensibility analysis of saturation line depth to tailings dam stability with index satisfaction degree[J]. China Safety Science Journal, 2013, 23(10):93-99.
[7]	李小军, 梅国栋, 苏军, 等. 灰色理论模型在尾矿库浸润线监测预测的应用[J]. 中国矿业, 2021, 30(增1):130-133.
	LI Xiaojun, MEI Guodong, SHU Jun, et al. Application of grey theory model in monitoring and prediction of tailings pond infiltration line[J]. China Mining Magazine, 2021, 30(S1):130-133.
[8]	许同乐, 王营博, 孟祥川, 等. 改进的灰色神经网络预测方法[J]. 北京邮电大学学报, 2018, 41(6):52-57,64. doi: 10.13190/j.jbupt.2018-072
	XU Tongle, WANG Yingbo, MENG Xiangchuan, et al. Improved grey neural networks prediction method[J]. Journal of Beijing University of Posts and Telecommuncations, 2018, 41(6):52-57,64.
[9]	戴健非, 杨鹏, 诸利一, 等. 集成PCA和LSTM神经网络的浸润线预测方法[J]. 中国安全科学学报, 2020, 30(3):94-101. doi: 10.16265/j.cnki.issn1003-3033.2020.03.015
	DAI Jianfei, YANG Peng, ZHU Liyi, et al. A PCA-LSTM neural network-integrated method for phreatic line prediction[J]. ChinaSafety Science Journal, 2020, 30(3):94-101.
[10]	刘传立, 刘小生, 李妍妍. 基于GEP的金属矿尾矿坝变形预测模型研究[J]. 有色金属科学与工程, 2013, 4(6):63-68.
	LIU Chuanli, LIU Xiaosheng, LI Yanyan. Deformation prediction model of metal mine tailings dam based on GEP[J]. Nonferrous Metals Science and Engineering, 2013, 4(6):63-68.
[11]	杜年春, 粟闯. 基于时间序列分析的浸润线预测分析[J]. 电子技术与软件工程, 2013(23):204-205.
[12]	随晓丹, 罗周全, 秦亚光, 等. 基于小波分解的尾矿坝浸润线预测方法研究[J]. 黄金科学技术, 2019, 27(1):137-143.
	SUI Xiaodan, LUO Zhouquan, QIN Yaguang, et al. Study onprediction method of seepage line of tailings dam based on wavelet decomposition[J]. Gold Science and Technology, 2019, 27(1):137-143.
[13]	邹伟霞, 李慧敏. 基于灰色模型GM(1,1)的尾矿坝浸润线预测[J]. 现代矿业, 2011, 27(11):314-316.
	ZOU Weixia, LI Huimin. Saturation line prediction oftailings dam based on grey model (1,1)[J]. Modern Mining, 2011, 27(11): 314-316.
[14]	邱俊博, 胡军. 基于ELM的尾矿坝浸润线预测[J]. 有色金属工程, 2021, 11(2):103-109.
	QIU Junbo, HU Jun. Prediction of phreatic line of tailings dam based on ELM[J]. Nonferrous Metals Engineering, 2021, 11(2): 103-109.
[15]	李春民, 王云海, 张兴凯, 等. 尾矿坝浸润线序列的支持向量机预测研究[J]. 金属矿山, 2010(12):18-21.
	LI Chunmin, WANG Yunhai, ZHANG Xingkai, et al. Prediction of infiltration route series intailings dam by the support vector machine[J]. Metal Mine, 2010(12):18-21.
[16]	SALAMON J, BELLO J. Deepconvolutional neural networks and data augmentation for environmental sound classification[J]. IEEE Signal Processing Letters, 2017, 24(3):279-283. doi: 10.1109/LSP.2017.2657381
[17]	CHO K, MERRIENBOER B V, GULCEHRE C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation[J]. Computer Science, 2014:1 724-1 734.
[18]	LOSHCHILOV I, HUTTER F. Fixing weight decay regularization in ADAM[C]. Proceedings in the International Conference on Learning Representations, 2018: 1-4.
[19]	GULCEHRE C, MOCZULSKI M, DENIL M, et al. Noisy Activation Functions[C]. Proceedings in the International Conference on Machine Learning, 2016:3 059-3 068.
[20]	CLEVERT D, UNTERTHINER T, HOCHREITER S. Fast andaccurate deep network learning by exponential linear units (ELUs)[J]. Computer Science, 2015: 1-14.
[21]	HOCHREITER S. Thevanishing gradient problem during learning recurrent neural nets and problem solutions[J]. International Journal of Uncertainty, 1998, 6(2): 107-116.
[22]	KRAUSE P, BOYLE D P, BÄSE F. Comparison of different efficiency criteria for hydrological model assessment[J]. Advances in Geosciences, 2005, 5(4): 89-97. doi: 10.5194/adgeo-5-89-2005
[23]	唐成龙, 谌颃, 唐海春, 等. 大数据背景下数据预处理方法研究运用[J]. 信息记录材料, 2021, 22(9):199-200.

监测点	监测项目描述
GT1、GT2	监测干滩长度,干滩长度是指从滩顶到库区水边的水平距离,可以反映尾矿库的安全运行情况
KSW	库水位监测,能反映尾矿库的排放能力,确保选矿厂的生产用水
JYL	降雨量监测,尾矿库的安全稳定性很大程度上受降雨等环境变化影响
WBWY-1-21、 WBWY-2-21、 WBWY-3-21 NBWY2-21-1、 NBWY2-21-2、 NBWY2-21-3	坝体位移监测,通过内、外部位移的变化可以捕捉坝体变形的变量和趋势,常用于尾矿库早期预警
JRX-1-21、 JRX-2-21、 JRX-3-21	浸润线监测,浸润线是坝体渗透水的顶部水面线,与坝体的安全稳定性息息相关

卷积层	MAPE/%	MAE	MSE	RMSE	R²
1	2.036 892	0.008 156	0.000 113	0.010 630	0.995 809
2	1.840 710	0.008 271	0.000 121	0.011 020	0.995 497
3	1.065 126	0.005 044	0.000 064	0.007 978	0.997 640

卷积核个数	MAPE/%	MAE	MSE	RMSE	R²
8,16,32	3.568 420	0.018 114	0.000 680	0.026 080	0.974 77
16,32,64	1.526 020	0.006 461	0.000 070	0.008 367	0.997 404
32,64,128	2.112 406	0.007 768	0.000 123	0.011 110	0.995 423
64,128,256	2.193 322	0.008 283	0.000 128	0.011 317	0.995 251

GRU层数	MAPE/%	MAE	MSE	RMSE	R²
1	2.474 627	0.010 081	0.000 154	0.012 426	0.994 275
2	2.247 059	0.008 414	0.000 129	0.011 338	0.995 233
3	1.590 019	0.007 086	0.000 085	0.009 234	0.996 838
4	2.635 393	0.012 423	0.000 239	0.015 448	0.992 482

GRU层神经元/个	MAPE/%	MAE	MSE	RMSE	R²
16	2.357 062	0.009 037	0.000 131	0.011 458	0.995 132
32	1.815 581	0.007 592	0.000 123	0.011 072	0.992 482
64	1.767 266	0.006 059	0.000 067	0.008 204	0.997 504
128	2.575 169	0.009 158	0.000 150	0.012 252	0.994 433

基于CNN-aGRU融合模型的尾矿坝浸润线预测方法

Prediction method of saturation line of tailings dam based on CNN-aGRU fusion model

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 23

相关文章 15

编辑推荐

Metrics

本文评价

参数	描述	值
Batch_size	每批过神经网络的大小	10
Epochs	神经网络训练的轮次	400
Optimizer	优化器,计算更新模型参数	AdamW
Output	全连接层输出数	1
Learning rate	初始学习率	0.001
Dropout	丢弃率	0.3

模型	MAPE	MAE	MSE	RMSE	R²
LSTM	0.017 813 97	0.008 535	0.000 134	0.011 597	0.995 013
GRU	0.028 554 64	0.012 445	0.000 304	0.017 442	0.988 719
CNN-aGRU2	0.026 491 09	0.011 367	0.000 203	0.014 239	0.992 452
CNN-aGRU	0.013 915 62	0.005 432	0.000 045	0.006 702	0.998 334

[1]	张术琳, 张亚楠, 田超, 严翔, 鲁义, 施式亮. 基于CNN的化工园区火灾火焰图像识别研究[J]. 中国安全科学学报, 2024, 34(1): 179-186.
[2]	张琦, 吉日格勒. 基于卷积神经网络的煤泥水外溢检测算法[J]. 中国安全科学学报, 2023, 33(S1): 192-195.
[3]	王瑶涵, 宋泽阳, 张利冬. 基于卷积神经网络的安全标识分类算法研究[J]. 中国安全科学学报, 2023, 33(S1): 263-269.
[4]	李善梅, 周相志. 基于S2S-CNN-GRU的机场离港航班延误预测[J]. 中国安全科学学报, 2023, 33(8): 93-100.
[5]	冉瑞生, 李进, 董殊宏. 基于Swin Transformer的弱监督人群计数研究[J]. 中国安全科学学报, 2023, 33(3): 111-117.
[6]	刘迪, 卢才武, 连民杰, 顾清华, 景莹. 基于贝叶斯决策的尾矿坝干滩长度预警研究[J]. 中国安全科学学报, 2022, 32(S1): 120-126.
[7]	曾纪涵, 章光, 吴浩, 胡少华. 改进POT模型下尾矿坝综合预警值确定方法[J]. 中国安全科学学报, 2022, 32(5): 134-139.
[8]	蒋水华, 曾绍慧, 黄劲松, 姚池. 尾矿材料渗透系数序贯概率反演分析[J]. 中国安全科学学报, 2020, 30(6): 158-165.
[9]	赵欣然, 张琪, 王卫东, 徐志强. 可燃性粉尘云的图像检测方法[J]. 中国安全科学学报, 2020, 30(4): 8-13.
[10]	武立功, 肖利兴, 刘晓峰, 姚池, 姜清辉, 周创兵. 尾砂粒径对尾矿坝漫顶溃坝的影响[J]. 中国安全科学学报, 2020, 30(4): 160-165.
[11]	戴健非, 杨鹏, 诸利一, 郭攀, 贯怀光. 集成PCA和LSTM神经网络的浸润线预测方法[J]. 中国安全科学学报, 2020, 30(3): 94-101.
[12]	刘克辉, 吴冬明, 蔡海, 敬小非, 李开放, 康钦容. 沟槽弯度对尾矿库溃决矿浆运动特性影响研究[J]. 中国安全科学学报, 2020, 30(2): 54-59.
[13]	熊若鑫, 宋元斌, 王宇轩, 段彦娟. 基于CNN的3D姿势估计在建筑工人行为分析中的应用[J]. 中国安全科学学报, 2019, 29(7): 64-69.
[14]	李辉, 易富, 张佳. 基于因素空间的尾矿坝稳定性综合评价[J]. 中国安全科学学报, 2019, 29(12): 28-34.
[15]	徐丹, 代勇, 纪军红. 基于卷积神经网络的驾驶人行为识别方法研究[J]. 中国安全科学学报, 2019, 29(10): 12-17.