基于XLNET模型的开阳磷矿成矿条件相关地质实体识别与应用

彭彬; 田宜平; 曾斌; 吴雪超; 吴文明

doi:10.19509/j.cnki.dzkq.tb20230543

基于XLNET模型的开阳磷矿成矿条件相关地质实体识别与应用

doi: 10.19509/j.cnki.dzkq.tb20230543

彭彬^1a,,
田宜平^{1a, 1b, 1c, 2, 3, ,},
曾斌^1a,
吴雪超^{1a, 3, 4},
吴文明^{2, 3, 5}

1a.
中国地质大学(武汉)计算机学院, 武汉 430078
1b.
中国地质大学(武汉)生物地质与环境地质国家重点实验室, 武汉 430078
1c.
中国地质大学(武汉)智能地学信息处理湖北省重点实验室, 武汉 430078
2.
自然资源部基岩区矿产资源勘查工程技术创新中心, 贵阳 550081
3.
贵州省战略矿产智慧勘查重点实验室, 贵阳 550081
4.
武汉地大坤迪科技有限公司, 武汉 430200
5.
贵州省地质矿产勘查开发局一○五地质大队, 贵阳 550018

基金项目:

中央引导地方科技发展资金项目“锰矿资源深部预测勘查技术研发基地” 黔科中引地[2021]4027

“贵州磷、锰、铝优势资源成矿规律与快速高效智慧化勘查技术研究及示范项目” 黔科合战略找矿[2022]ZD003

“贵州磷、锰、铝优势资源成矿规律与快速高效智慧化勘查技术研究及示范项目” [2022]ZD004

智能地学信息处理湖北省重点实验室2022年度开放研究课题 KLIGIP-2022-B05

2021年度生物地质与环境地质国家重点实验室自主课题 128-GKZ21Y647

详细信息

作者简介:
彭彬, E-mail: 2499029434@qq.com

通讯作者:
田宜平, E-mail: yptian@cug.edu.cn

中图分类号: P619.21
计量
- 文章访问数: 242
- PDF下载量: 66
- 被引次数: 0
出版历程
- 收稿日期: 2023-09-25
- 录用日期: 2024-03-21
- 修回日期: 2023-11-23

Recognition and application of geological entities related to ore-forming conditions in the Kaiyang phosphate mine based on the XLNET model

PENG Bin^{1a
,},
TIAN Yiping^{1a, 1b, 1c, 2, 3
, ,},
ZENG Bin^1a,
WU Xuechao^{1a, 3, 4},
WU Wenming^{2, 3, 5}

1a.
School of Computer Science, China University of Geosciences(Wuhan), Wuhan 430078, China
1b.
State Key Laboratory of Biogeology and Environmental Geology, China University of Geosciences(Wuhan), Wuhan 430078, China
1c.
Hubei Key Laboratory of Intelligent Geo-Information Processing, China University of Geosciences(Wuhan), Wuhan 430078, China
2.
Engineering Technology Innovation Center of Mineral Resources Explorations in Bedrock Zones, Ministry of Natural Resources, Guiyang 550081, China
3.
Guizhou Key Laboratory for Strategic Mineral Intelligent Exploration, Guiyang 550081, China
4.
Wuhan Dida Kundi Science and Technology Co., Ltd., Wuhan 430200, China
5.
Geology Team 105, Bureau of Geology and Mineral Exploration and Development, Guizhou Province, Guiyang 550081, China

More Information

Author Bio:
PENG Bin, E-mail: 2499029434@qq.com

Corresponding author: TIAN Yiping, E-mail: yptian@cug.edu.cn

摘要

摘要:
随着磷矿找矿难度越来越大, 地质勘探成果报告也愈来愈多, 通过人工识别海量文档中与磷矿成矿相关地质信息耗时低效, 无法满足知识共享传播和地质报告智能管理的需求。为快速获得磷矿地质文档报告中隐藏的成矿地质知识, 基于XLNET模型建立了磷矿成矿地质实体自动识别的方法。首先对实体进行BIO标注建立地质实体字典, 利用XLNET作为底层预处理模型学习句子双向语义; 然后使用BILSTM-Attention-CRF模型实现文本多标签的智能分类; 最后通过定位磷矿实体在报告中的分布位置大致推测该处磷矿成矿条件和成矿模式。将该模型与其余3种模型比较得出结果, 该模型识别的准确率(P)、召回率(R)及F1值都接近了90%, 较前3种模型分别调高了2%, 5%, 6%。该研究为开阳磷矿地质研究人员提供了更加高效的地质实体自动识别的方法。
- 地质实体识别 /
- XLNET-BILSTM-Attention-CRF /
- 磷矿成矿模式 /
- 预训练模型 /
- 序列标注
Abstract: Objective
With increasing difficulty in phosphate ore prospecting, there are an increasing number of geological exploration reports. The manual recognition of geological information related to phosphate rock mineralization in massive documents is time-consuming and inefficient. It cannot meet the needs of knowledge sharing, dissemination and intelligent management of geological reports.
Methods
To quickly obtain the ore-forming geological knowledge hidden in the phosphate ore reports, this work intends to establish an automatic recognition method for ore-forming geological entities based on the extreme learning machine network(XLNET) model. First, BIO labelling of entities was carried out to establish a geological entity dictionary, and XLNET was used as the underlying preprocessing model to learn the bidirectional semantics of sentences. Then, the BILSTM-Attention-CRF(bidirectional long short term memory(BILSTM)-self attention layer(Attention)-conditional random field(CRF)) model was used to realize intelligent classification of multiple text labels. Finally, the ore-forming conditions and ore-forming model of phosphate ore in the reports were roughly predicted by locating the distribution position of phosphate ore entities in the report.
Results
Comparing this model with the other three models, these results show that the accuracy rate, recall rate and F1 value of this model are close to 90%, which are 2%, 5% and 6% higher than those of the previous three models, respectively.
Conclusion
This study provides a more efficient method for automatic geological entity recognition for geological researchers in the Kaiyang phosphate mine.
注释:

所有作者声明不存在利益冲突。

HTML全文

图 1 Transformer-XL循环机制

Figure 1. Transformer-XL loop mechanism

下载: 全尺寸图片幻灯片

图 2 基于文本的地质实体识别方法技术路线

Figure 2. Technical flowchart of the proposed text-based geological entity recognition

下载: 全尺寸图片幻灯片

图 3 模型架构及算法流程图

Figure 3. Model architecture and algorithm flow chart

下载: 全尺寸图片幻灯片

图 4 实体类别数量分布(符号含义见表 1，2，下同)

Figure 4. Distribution of the number of entity categories

下载: 全尺寸图片幻灯片

图 5 模型损失值变化趋势

Figure 5. Change trend of the model loss value

下载: 全尺寸图片幻灯片

图 6 不同实体识别热图

P.准确率；R.召回率

Figure 6. Different entity recognition hot renderings

下载: 全尺寸图片幻灯片

图 7 交互式磷矿地质实体识别平台

Figure 7. Interactive phosphate geological entity recognition platform

下载: 全尺寸图片幻灯片

图 8 实体和属性类型知识图谱

Figure 8. Knowledge graph of the entity and attribute types

下载: 全尺寸图片幻灯片

表 1 磷矿实体标签

Table 1. Phosphate ore entity labels

标签	实体类型
KQ	矿区
KC	矿床
KD	矿段
KT	矿体

下载: 导出CSV

表 2 磷矿实体属性标签

Table 2. Phosphate ore entity attribute labels

标签	属性类型
LX	类型
GM	规模
SJ	时间
KJ	空间
DC	地层(或断层)
KKGZ	控矿构造
CZ	产状
XT	形态
SBLX	蚀变类型
DLGM	定量规模
FYCD	发育程度

下载: 导出CSV

表 3 磷矿实体语义关系标签

Table 3. Phosphate ore entity semantic relationship labels

标签	实体类型
R1	(时间) 4种实体和时间
R2	(包含关系) 4种实体之间
R3	(类型与规模) 4种实体和类型与规模
R4	(空间) 4种实体和空间
R5	(基本特征) 4种实体和产状、形态、蚀变类型、定量规模、发育程度
R6	(赋存部位) 4种实体和地层(或断层)、控矿构造

下载: 导出CSV

表 4 磷矿实体属性关系标注示例

Table 4. Phosphate ore entity attribute relationship annotation example

实体	实体或属性	关系类型
洋水矿区(KQ)	西翼(KJ)	R4
洋水矿区(KQ)	浅-滨海相沉积磷块岩矿床(KC)	R2
洋水矿区(KQ)	隆起(KKGZ)	R6
洋水矿区(KQ)	深度30~500 m(DLGM)	R5
洋水矿区(KQ)	南沱冰期(SJ)	R1
洋水矿区(KQ)	海侵(SBLX)	R5
洋水矿区(KQ)	层状(XT)	R5
洋水矿区(KQ)	Ⅴ号矿体(KT)	R2
洋水矿区(KQ)	Ⅵ号矿体(KT)	R2
Ⅴ号矿体(KT)	F₁₁(DC)	R6
Ⅴ号矿体(KT)	下盘(KJ)	R4
…	…	…

下载: 导出CSV

表 5 实验环境参数

Table 5. Experimental environment settings

操作系统	CPU	内存	硬盘容量	GPU
Linux	AMD EPYC 7543	80GB	80GB	A40

下载: 导出CSV

表 6 模型参数列表

Table 6. Model parameter list

LSTM隐层维数	最长序列长度	批次大小	AdamW学习率	Dropout
1 024	128	36	3×10^-5	0.5

下载: 导出CSV

表 7 不同模型地质识别效果

Table 7. Geological recognition effect of different models

模型	准确率P/%	召回率R/%	F1值/ %
迭代400次
BiLSTM-CRF	76.41	73.74	75.05
BERT-BiLSTM-CRF	79.52	78.01	78.76
XLNET-BiLSTM-CRF	82.38	83.29	82.32
XLNET-BiLSTM-Attention-CRF	85.75	86.91	86.32
迭代800次
BiLSTM-CRF	84.59	83.11	83.85
BERT-BiLSTM-CRF	85.43	84.23	84.83
XLNET-BiLSTM-CRF	87.51	88.73	88.12
XLNET-BiLSTM-Attention-CRF	89.03	90.86	89.94

下载: 导出CSV

表 8 地质实体识别结果

Table 8. Geological entity recognition results

实例	原文信息	人工标注实体	模型识别实体
例一	两岔河矿段位于洋水背斜西翼中北部	KD：“两岔河矿段”, KJ；“洋水背斜西翼中北部”	KD: “两岔河矿段”；KJ: “洋水背斜西翼中北部”
例二	目前该矿脉已探明Ⅰ号、Ⅱ号、Ⅲ号矿体	KT：“Ⅰ号”“Ⅱ号”“Ⅲ号”	KT：“Ⅰ号”“Ⅱ号”“Ⅲ号”矿体
例三	工作区出露地层主要为青白口系鹅家坳组	DC：“青白口系鹅家坳组”	DC：“青白口系鹅家坳组”
例四	贵州省开阳磷矿洋水矿区两岔河矿段(南段)磷矿勘探	KT：“磷矿”; KQ；“洋水矿区”; KD；“两岔河矿段”	KT：“磷矿”；KQ：“洋水矿区”；KD：“两岔河矿段”

下载: 导出CSV

表 9 模型在扩大训练集规模之后的地质实体识别结果

Table 9. Geological entity recognition results of the model after expanding the training set

训练集规模	P/%	R/%	F1值/%
5万余字	91.36	94.67	92.99
7万余字	93.42	95.36	94.38

下载: 导出CSV

参考文献(19)

[1]	JIN G Z, LIN Y F, XIAO G M, et al. SEM+: Tool for discovering concept mapping in earth science related domain[J]. Earth Science Informatics, 2015, 8(1): 95-102. doi: 10.1007/s12145-014-0203-1
[2]	LIANG W, LEI X, CHAO L L, et al. A knowledge-driven geospatially enabled framework for geological big data[J]. ISPRS International Journal of Geo-Information, 2017, 6(6): 166-166. doi: 10.3390/ijgi6060166
[3]	QIN J Q, XIE Z, WU L, et al. BiLSTM-CRF for geological named entity recognition from the geoscience literature[J]. Earth Science Informatics, 2019, 12(4): 565-579. doi: 10.1007/s12145-019-00390-3
[4]	QIN J Q, ZHONG X, LIANG W, et al. GNER: A generative model for geological named entity recognition without labeled data using deep learning[J]. Earth and Space Science, 2019, 6(6): 931-946. doi: 10.1029/2019EA000610
[5]	储德平, 万波, 李红, 等. 基于ELMO-CNN-BiLSTM-CRF模型的地质实体识别[J]. 地球科学, 2021, 46(8): 3039-3048. https://www.cnki.com.cn/Article/CJFDTOTAL-DQKX202108028.htm CHU D P, WAN B, LI H, et al. Geological entity recognition based on ELMO-CNN-BiLSTM-CRF model[J]. Earth Science, 2021, 46(8): 3039-3048. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-DQKX202108028.htm
[6]	邱芹军. 基于地质报告文本的时空及主题提取关键技术研究[D]. 武汉: 中国地质大学(武汉), 2020. QIU Q J. Rsearch on the key technologies of spatio-temporal and topic extraction based on geological report text[D]. Wuhan: China University of Geosciences (Wuhan), 2020. (in Chinese with English abstract)
[7]	吴冲龙, 刘刚, 周琦, 等. 地质科学大数据统合应用的基本问题[J]. 地质科技通报, 2020, 39(4): 1-11. doi: 10.19509/j.cnki.dzkq.2020.0401 WU C L, LIU G, ZHOU Q, et al. Fundamental problems of intergrated application of big data in geoscience[j]. Bulletin of Geological Science and Technology, 2020, 39(4): 1-11. (in Chinese with English abstract) doi: 10.19509/j.cnki.dzkq.2020.0401
[8]	张雪英, 叶鹏, 王曙, 等. 基于深度信念网络的地质实体识别方法[J]. 岩石学报, 2018, 34(2): 343-351. https://www.cnki.com.cn/Article/CJFDTOTAL-YSXB201802011.htm ZHANG X Y, YE P, WANG S, et al. Geological entity recognition method based on deep belief networks[J]. Acta Petrologica Sinica, 2018, 34(2): 343-351. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-YSXB201802011.htm
[9]	马凯. 地质大数据表示与关联关键技术研究[D]. 武汉: 中国地质大学(武汉), 2018. MA K. Research on the key technologies of geological big data representation and association[D]. Wuhan: China University of Geosciences (Wuhan), 2018. (in Chinese with English abstract)
[10]	谢雪景, 谢忠, 马凯, 等. 结合BERT与BiGRU-Attention-CRF模型的地质命名实体识别[J]. 地质通报, 2023, 42(5): 846-855. https://www.cnki.com.cn/Article/CJFDTOTAL-ZQYD202305014.htm XIE X J, XIE Z, MA K, et al. Geological named entity recognition combined BERT and BiGRU-Attention-CRF model[J]. Geological Bulletin of China, 2023, 42(5): 846-855. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-ZQYD202305014.htm
[11]	张春菊, 张磊, 陈玉冰, 等. 基于BERT的交互式地质实体标注语料库构建方法[J]. 地理与地理信息科学, 2022, 38(4): 7-12. https://www.cnki.com.cn/Article/CJFDTOTAL-DLGT202204002.htm ZHANG C J, ZHANG L, CHEN Y B, et al. BERT-based interactive geological entity annotation corpus construction method[J]. Geography & Geographic Information Science, 2022, 38(4): 7-12. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-DLGT202204002.htm
[12]	王刘坤, 李功权. 基于GeoERNIE-BiLSTM-Attention-CRF模型的地质命名实体识别[J]. 地质科学, 2023, 58(3): 1164-1177. https://www.cnki.com.cn/Article/CJFDTOTAL-DZKX202303022.htm WANG L K, LI G Q. Geological named entity recognition based on GeoERNIE-BiLSTM-Attention-CRF model[J]. Chinese Journal of Geology, 2023, 58(3): 1164-1177. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-DZKX202303022.htm
[13]	ZHAO H, HUANG C N, MU L, et al. An improved Chinese word segmentation system with conditional random field[C]//Anon. Proceedings of the fifth sighan workshop on Chinese language processing. 2006: 108-117.
[14]	梁坤萍, 程国繁, 覃庆炎, 等. 贵州织金新华磷矿区风化磷块岩形成条件及风化淋滤富集机制初步研究[J]. 地质科技通报, 2022, 41(4): 172-183. doi: 10.19509/j.cnki.dzkq.2022.0110 LIANG K P, CHENG G F, QING Q Y, et al. A preliminary study on the formation conditions and weathering leaching enrichment mechanism of secondary phosphorite in the Xinhua phosphate mining area, Zhijin, Guizhou[J]. Bulletin of Geological Science and Technology, 2022, 41(4): 172-183. (in Chinese with English abstract) doi: 10.19509/j.cnki.dzkq.2022.0110
[15]	程国繁, 何英. 贵州册亨板其风化型磷矿成矿条件与成矿模式[J]. 矿物学报, 2016, 36(2): 189-197. https://www.cnki.com.cn/Article/CJFDTOTAL-KWXB201602004.htm CHEN G F, HE Y. A preliminary study on ore-forming conditions and its model for Banqi secondary phosphate deposit, Ceheng County, Guizhou Province, China[J]. Acta Mineralogica Sinica, 2016, 36(2): 189-197. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-KWXB201602004.htm
[16]	张亚冠, 杜远生, 陈国勇, 等. 富磷矿三阶段动态成矿模式: 黔中开阳式高品位磷矿成矿机制[J]. 古地理学报, 2019, 21(2): 351-368. https://www.cnki.com.cn/Article/CJFDTOTAL-GDLX201902011.htm ZHANG Y G, DU Y S, CHEN G Y, et al. Three stages dynamic mineralization model of the phosphate-rich deposits: Mineralization mechanism of the Kaiyang-type high-grade phosphorite in central Guizhou Province[J]. Journal of Palaeogeography, 2019, 21(2): 351-368. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-GDLX201902011.htm
[17]	姚贵斌, 张起贵. 基于XLnet语言模型的中文命名实体识别[J]. 计算机工程与应用, 2021, 57(18): 156-162. https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG202118019.htm YAO G B, ZHANG Q G. Chinese named entity recognition based on XLnet language model[J]. Computer Engineering and Applications, 2021, 57(18): 156-162. (in Chinese with English abstract) https://www.cnki.com.cn/Article/CJFDTOTAL-JSGG202118019.htm
[18]	WANG C B, LI Y J, CHEN J G, et al. Named entity annotation schema for geological literature mining in the domain of porphyry copper deposits[J]. Ore Geology Reviews, 2022, 152: 105243.
[19]	王龙辉, 剡鹏兵, 焦养泉, 等. 鄂尔多斯盆地北部下白垩统铀成矿模式[J]. 地质科技通报, 2023, 42(3): 222-233. doi: 10.19509/j.cnki.dzkq.2022.0096 WANG L H, YAN P B, JIAO Y Q, et al. Uranium metallogenic model of the Lower Cretaceous in the northern Ordos Basin[J]. Bulletin of Geological Science and Technology, 2023, 42(3): 222-233. (in Chinese with English abstract) doi: 10.19509/j.cnki.dzkq.2022.0096