基于数据预处理和改进麻雀算法的岩爆预测

张鼎; 周宗红

doi:10.11858/gywlxb.20240964

基于数据预处理和改进麻雀算法的岩爆预测

doi: 10.11858/gywlxb.20240964

张鼎,
周宗红^*, ,

昆明理工大学国土资源工程学院, 云南昆明　650093

基金项目: 国家自然科学基金（52264019）

详细信息

作者简介:
张　鼎（1998－），男，硕士研究生，主要从事采矿与岩石力学研究. E-mail：1441460925@qq.com

通讯作者:
周宗红（1967－），男，博士，教授，主要从事采矿与岩石力学研究. E-mail：zhou20051001@163.com

中图分类号: O382; TD311; O521.9
计量
- 文章访问数: 151
- HTML全文浏览量: 42
- PDF下载量: 16
出版历程
- 收稿日期: 2024-12-25
- 修回日期: 2025-03-04
- 录用日期: 2025-04-25
- 网络出版日期: 2025-03-11
- 刊出日期: 2025-07-07

Rock Burst Prediction Based on Data Preprocessing and Improved Sparrow Algorithm

ZHANG Ding,
ZHOU Zonghong^{*
, ,}

School of Land and Resources Engineering, Kunming University of Science and Technology, Kunming 650093, Yunnan, China

摘要

摘要: 为解决机器学习岩爆预测中存在离群样本、样本不均衡、麻雀搜索算法易陷入局部最优等问题，从数据预处理和算法改进2个角度建立岩爆预测模型。首先，基于岩性条件和应力条件，选取围岩最大切向应力、抗压强度、抗拉强度和弹性能量指数作为特征指标，采用3种机器学习算法，结合5折交叉验证方法构建预测模型。在数据预处理阶段，收集了174组国内外岩爆案例建立数据库，针对离群样本，引入局部离群因子（LOF）算法，根据岩爆等级逐级检测并剔除离群样本；针对样本不均衡问题，引入自适应过采样方法（ADASYN）增加少数类样本数目。采用3种混合策略改进麻雀搜索算法，利用改进的麻雀搜索算法（ISSA）对极限梯度提升树（XGBoost）、随机森林（RF）、多层感知机（MLP）3种机器学习算法参数寻优，分析准确率、精确率等多个评价指标，对模型进行有效性验证。结果表明，新构建的最优模型ISSA-XGBoost的准确率达到了94.12%，具有较高的预测准确率。此外，对4种特征指标进行特征重要性分析，确定了围岩最大切向应力是最重要特征。
- 岩爆预测 /
- 数据预处理 /
- 算法改进 /
- 特征重要性 /
- 机器学习
Abstract: To solve the problems of outlier samples, imbalanced samples, and local optimal of sparrow search algorithm in machine learning rockburst prediction, this paper established a rockburst prediction model from two perspectives of data preprocessing and algorithm improvement. First, based on lithology conditions and stress conditions, selected the maximum tangential stress, compressive strength, tensile strength and elastic energy index of surrounding rock as the characteristic indexes, and used three kinds of machine learning algorithms combined with 5-fold cross-validation method to construct the prediction model. In the data pre-processing stage, collected 174 groups of domestic and international rock burst cases to establish a database; for outlier samples, introduced the local outlier factor (LOF) algorithm to detect and eliminate outlier samples step by step according to the rock burst class; for sample imbalance, the adaptive synthetic sampling method (ADASYN) was introduced to increase the number of minority class samples. Three hybrid strategies were employed to improve sparrow search algorithm (ISSA) was used to optimize the parameters of three machine learning algorithms, namely limit gradient lift tree (XGBoost), random forest (RF) and multi-layer perceptron (MLP). Multiple evaluation indexes such as accuracy rate and precision rate were analyzed and discussed to verify the effectiveness of the model. The results show that the accuracy of the newly constructed optimal model, ISA-XGBoost, reaches 94.12%, indicating high prediction accuracy. In addition to the feature importance analysis of the four feature indexes, it was determined that the maximum tangential stress of the surrounding rock is the most important feature.
- rockburst prediction /
- data preprocessing /
- algorithm improvement /
- feature importance /
- machine learning

HTML全文

图 1 原始数据统计学关系

Figure 1. Statistical relationship of raw data

下载: 全尺寸图片幻灯片

图 2 数据集中岩爆等级的分布

Figure 2. Distribution of rock burst grade in data set

下载: 全尺寸图片幻灯片

图 3 离群样本散点图（σ_θmax与σ_c）

Figure 3. Scatter plot of outlier samples (σ_θmax与σ_c)

下载: 全尺寸图片幻灯片

图 4 不同测试函数收敛曲线

Figure 4. Convergence curves of different test functions

下载: 全尺寸图片幻灯片

图 5 交叉验证和目标函数计算

Figure 5. Cross-validation and objective function calculation

下载: 全尺寸图片幻灯片

图 6 岩爆预测模型流程图

Figure 6. Flow chart of rock burst prediction model

下载: 全尺寸图片幻灯片

图 7 SHAP特征摘要

Figure 7. Summary of SHAP features

下载: 全尺寸图片幻灯片

图 8 特征重要性分析

Figure 8. Feature importance analysis

下载: 全尺寸图片幻灯片

表 1 岩爆经典案例

Table 1. Classic rock burst cases

Sample	σ_θmax/MPa	σ_c/MPa	σ_t/MPa	λ_wet	Rock burst grade
1	18.80	178.00	7.40	5.70	Ⅰ
2	31.05	147.85	3.00	11.96	Ⅲ
3	91.43	157.63	6.27	11.96	Ⅳ
4	61.91	92.40	5.43	8.28	Ⅲ
5	58.20	83.60	5.90	2.60	Ⅳ
6	60.00	136.79	2.12	10.42	Ⅱ
7	44.40	120.00	5.10	5.00	Ⅱ
…	…	…	…	…	…
170	61.46	135.67	9.02	11.20	Ⅳ
172	110.55	106.20	9.59	12.51	Ⅳ
172	41.90	143.01	6.68	4.30	Ⅱ
173	109.40	190.00	6.10	6.90	Ⅲ
174	48.00	59.00	5.23	0.88	Ⅰ

下载: 导出CSV

表 2 岩爆分类等级特征描述

Table 2. Characteristics of rock burst grade

Rock burst grade	Feature description
Rock burst grade	Motion feature	Sweep depth/m	Acoustic signature
Ⅰ	No rockburst activities	0	No sound
Ⅱ	Rock deforms, cracks or spalls, but no ejection phenomenon occurs	<0.5	Weak sound
Ⅲ	Rock deforms and fractures with a considerable amount of chip ejecting, loosening and sudden destruction	0.5–1.0	Crisp crackling
Ⅳ	Rock deforms and fractures with a considerable amount of chip ejecting, loosening and sudden destruction	>1.0	Strong bursts and roaring sounds

下载: 导出CSV

表 3 岩爆分级标准

Table 3. Rock burst classification criteria

Rock burst grade	Rock burst evaluation index
Rock burst grade	σ_θmax/MPa	σ_c/MPa	σ_t/MPa	λ_wet
Ⅰ	0–24	0–80	0–5	0–2.0
Ⅱ	24–60	80–120	5–7	2.0–3.5
Ⅲ	60–126	120–180	7–9	3.5–5.0
Ⅳ	126–200	180–320	9–30	5.0–20.0

下载: 导出CSV

表 4 数据集统计参数

Table 4. Statistical parameters of data sets

Rock burst index	σ_θmax/MPa	σ_c/MPa	σ_t/MPa	λ_wet
Maximum value	221.00	200.72	22.60	17.13
Minimum value	3.80	15.50	0.70	0.88
Mean value	50.42	113.68	4.94	6.16
Standard deviation	29.42	38.51	2.88	3.57
Sample size	174	174	174	174

下载: 导出CSV

表 5 剔除离群样本

Table 5. Removes outlier samples

Sample No.	σ_θmax/MPa (outlier value)	Rock burst grade	Sample No.	σ_θmax/MPa (outlier value)	Rock burst grade
11	108.40	Ⅳ	95	75.60	Ⅲ
25	110.35	Ⅳ	96	59.77	Ⅳ
36	2.60	Ⅰ	122	119.69	Ⅳ
…	…	…

下载: 导出CSV

表 6 算法参数

Table 6. Algorithm parameters

PSO	GWO	GA	SSA	ISSA
c₁=c₂=2, w=0.9	a$\in $[0, 2]; r₁, r₂$\in $[0, 1]	P_c=0.8, P_m=0.05	P_D=20%, S_D=10%, S=0.8	P_D=20%, S_D=10%, S=0.8, w_min=1, w_max=3

下载: 导出CSV

表 7 测试函数

Table 7. Test functions

Function	Type	Value range	Optimal solution
F₁	Sphere	[–100, 100]	0
F₂	Schwefel′2.22	[–10, 10]	0
F₃	Schwefel′1.2	[–100, 100]	0
F₄	Quartic	[–1.28, 1.28]	0
F₅	Rastrigrin	[–5.12, 5.12]	0
F₆	Ackley	[–32, 32]	0
F₇	Griewing	[–600, 600]	0
F₈	Foxholes	[–65, 65]	1

下载: 导出CSV

表 8 ISSA与经典算法实验结果的对比

Table 8. ISSA experimental results compared with classical algorithms

Algorithm	Mean value
Algorithm	F₁	F₂	F₃	F₄	F₅	F₆	F₇	F₈
PSO	1.21×10³	1.84×10¹⁰	5.42×10¹⁰	6.20×10⁹	8.18×10¹⁰	6.52×10¹⁰	1.45×10⁸	2.21×10⁹
GWO	1.23×10³	2.19×10¹¹	2.63×10¹¹	1.93×10¹⁰	2.88×10⁶	6.40×10⁸	1.31×10¹⁰	2.10×10⁸
GA	8.28×10³	3.66×10¹¹	3.12×10¹¹	7.00×10¹¹	7.97×10⁸	6.17×10⁷	1.64×10¹¹	3.19×10¹¹
SSA	5.60×10⁻³	2.50×10⁻²	1.71×10⁻²	7.74×10⁻²	7.02×10⁻²	5.19×10⁻²	2.23×10⁻²	4.15×10⁻²
ISSA	6.11×10⁻⁴	1.39×10⁻²	1.08×10⁻²	1.75×10⁻²	0	7.67×10⁻³	0	2.87×10⁻²

Algorithm	Standard deviation
Algorithm	F₁	F₂	F₃	F₄	F₅	F₆	F₇	F₈
PSO	7.21×10³	4.11×10¹¹	1.21×10¹²	1.39×10¹¹	1.83×10¹²	1.46×10¹²	3.25×10⁹	4.94×10¹⁰
GWO	7.71×10³	4.88×10¹²	5.88×10¹²	3.52×10¹¹	4.54×10⁷	1.20×10¹⁰	2.89×10¹¹	4.08×10⁹
GA	1.41×10⁴	5.58×10¹²	6.80×10¹²	9.64×10¹²	1.43×10¹⁰	4.46×10⁸	2.14×10¹²	5.11×10¹²
SSA	6.32×10⁻²	3.01×10⁻¹	2.95×10⁻¹	9.80×10⁻¹	8.46×10⁻¹	7.58×10⁻¹	2.15×10⁻¹	5.78×10⁻¹
ISSA	8.20×10⁻³	1.47×10⁻¹	1.89×10⁻¹	2.75×10⁻¹	0	7.48×10⁻²	0	4.48×10⁻¹

下载: 导出CSV

表 9 数据预处理前后3种机器学习模型准确率

Table 9. Accuracy of three machine learning models before and after data preprocessing

Data class	Model accuracy/%
Data class	MLP	RF	XGBoost
Raw rock burst data set	55.17	60.34	62.07
Preprocessing rock burst data set	63.22	74.71	77.59

下载: 导出CSV

表 10 SSA改进前后模型预测结果的对比

Table 10. Comparison of model prediction results before and after SSA improvement

Algorithm	Rock burst grade	Precision rate	Recall rate	F₁ value	Sample size/group
SSA-MLP （Accuracy, 82.36%）	Ⅰ	1.00	1.00	1.00	13
	Ⅱ	0.75	0.75	0.75	12
	Ⅲ	0.80	0.68	0.74	12
	Ⅳ	0.75	0.86	0.80	14
SSA-RF （Accuracy, 84.31%）	Ⅰ	1.00	1.00	1.00	13
	Ⅱ	1.00	0.83	0.91	12
	Ⅲ	0.68	0.68	0.68	12
	Ⅳ	0.75	0.86	0.80	14
SSA-XGBoost （Accuracy, 87.00%）	Ⅰ	0.83	0.92	0.87	13
	Ⅱ	0.91	0.77	0.83	12
	Ⅲ	0.82	0.75	0.78	12
	Ⅳ	0.85	0.93	0.89	14
ISSA-MLP （Accuracy, 86.25%）	Ⅰ	1.00	1.00	1.00	13
	Ⅱ	0.83	0.83	0.83	12
	Ⅲ	0.80	0.68	0.74	12
	Ⅳ	0.81	0.93	0.87	14
ISSA-RF （Accuracy, 88.24%）	Ⅰ	1.00	0.92	0.96	13
	Ⅱ	0.84	0.83	0.83	12
	Ⅲ	0.79	0.91	0.85	12
	Ⅳ	0.92	0.86	0.89	14
ISSA-XGBoost （Accuracy, 94.12%）	Ⅰ	1.00	1.00	1.00	13
	Ⅱ	1.00	0.83	0.91	12
	Ⅲ	0.80	1.00	0.89	12
	Ⅳ	1.00	0.93	0.96	14

下载: 导出CSV

参考文献(21)

[1]	田睿. 基于机器学习的岩爆烈度等级预测模型研究与应用 [D]. 包头: 内蒙古科技大学, 2020. TIAN R. Research and application of rockburst intensity classification prediction model based on machine learning algorithms [D]. Baotou: Inner Mongolia University of Science and Technology, 2020.
[2]	刘剑, 周宗红, 刘军, 等. 基于主成分分析和改进Bayes判别的岩爆等级预测 [J]. 采矿与岩层控制工程学报, 2022, 4(5): 053014. doi: 10.13532/j.jmsce.cn10-1638/td.2022.05.004 LIU J, ZHOU Z H, LIU J, et al. Prediction of rockburst grade based on principal component analysis and improved Bayesian discriminant analysis [J]. Journal of Mining and Strata Control Engineering, 2022, 4(5): 053014. doi: 10.13532/j.jmsce.cn10-1638/td.2022.05.004
[3]	孙飞跃, 刘希亮, 郭佳奇, 等. 岩爆预测评估方法的动力数值分析 [J]. 应用力学学报, 2022, 39(1): 26–34. doi: 10.11776/j.issn.1000-4939.2022.01.004 SUN F Y, LIU X L, GUO J Q, et al. Dynamic numerical calculation analysis of rockburst prediction assessment methods [J]. Chinese Journal of Applied Mechanics, 2022, 39(1): 26–34. doi: 10.11776/j.issn.1000-4939.2022.01.004
[4]	曲宏略, 刘哲言, 杨龙, 等. 基于应力判据的隧道岩爆预测评估研究 [J]. 地下空间与工程学报, 2020, 16(Suppl 2): 934–938,956. QU H L, LIU Z Y, YANG L, et al. Prediction and evaluation of rock burst in tunnel based on stress criterion [J]. Chinese Journal of Underground Space and Engineering, 2020, 16(Suppl 2): 934–938,956.
[5]	DOU L M, CAI W, CAO A Y, et al. Comprehensive early warning of rock burst utilizing microseismic multi-parameter indices [J]. International Journal of Mining Science and Technology, 2018, 28(5): 767–774. doi: 10.1016/j.ijmst.2018.08.007
[6]	CAO A Y, JING G C, DING Y L, et al. Mining-induced static and dynamic loading rate effect on rock damage and acoustic emission characteristic under uniaxial compression [J]. Safety Science, 2019, 116: 86–96. doi: 10.1016/j.ssci.2019.03.003
[7]	温廷新, 陈依琳. 基于海林格距离和AHDPSO-ELM的岩爆烈度等级预测模型 [J]. 中国安全科学学报, 2022, 32(11): 38–46. doi: 10.16265/j.cnki.issn1003-3033.2022.11.1915 WEN T X, CHEN Y L. Prediction model of rockburst intensity grade based on Hellinger distance and AHDPSO-ELM [J]. China Safety Science Journal, 2022, 32(11): 38–46. doi: 10.16265/j.cnki.issn1003-3033.2022.11.1915
[8]	李明亮, 李克钢, 秦庆词, 等. 岩爆烈度等级预测的机器学习算法模型探讨及选择 [J]. 岩石力学与工程学报, 2021, 40(Suppl 1): 2806−2816. LI M L, LI K G, QIN Q C, et al. Discussion and selection of machine learning algorithm model for rockburst intensity grade prediction [J]. Chinese Journal of Rock Mechanics and Engineering, 201, 40(Suppl 1): 2806−2816.
[9]	侯克鹏, 包广拓, 孙华芬. 改进的MVO-GRNN神经网络岩爆预测模型研究 [J]. 安全与环境学报, 2024, 24(3): 923–932. doi: 10.13637/j.issn.1009-6094.2023.0341 HOU K P, BAO G T, SUN H F. Research on improved MVO-GRNN neural network rockburst prediction model [J]. Journal of Safety and Environment, 2024, 24(3): 923–932. doi: 10.13637/j.issn.1009-6094.2023.0341
[10]	满轲, 武立文, 刘晓丽, 等. 基于灰色关联分析和SSA-RF模型的岩爆等级预测 [J]. 金属矿山, 2023(5): 202–212. doi: 10.19614/j.cnki.jsks.202305021 MAN K, WU L W, LIU X L, et al. Rockburst grade prediction based on grey correlation analysis and SSA-RF model [J]. Metal Mine, 2023(5): 202–212. doi: 10.19614/j.cnki.jsks.202305021
[11]	高梅, 张成良, 张华超, 等. 基于SMOTEENN-CGAN-Stacking的岩爆烈度等级预测研究 [J]. 工程地质学报, 2024, 32(6): 2264–2276. doi: 10.13544/j.cnki.jeg.2024-0112 GAO M, ZHANG C L, ZHANG H C, et al. Rockburst intensity level prediction based on SMOTEENN-CGAN-Stakcing [J]. Journal of Engineering Geology, 2024, 32(6): 2264–2276. doi: 10.13544/j.cnki.jeg.2024-0112
[12]	苏焕博. 数据缺失和不均衡下的IDPSO-ELM岩爆烈度等级预测模型研究 [D]. 阜新: 辽宁工程技术大学, 2022. SU H B. Research on IDPSO-ELM rockburst intensity grade prediction model with missing and unbalanced data [D]. Fuxin: Liaoning Technical University, 2022.
[13]	ZHOU J, LI X, MITRI H S. Classification of rockburst in underground projects: comparison of ten supervised learning methods [J]. Journal of Computing in Civil Engineering, 2016, 30(5): 04016003. doi: 10.1061/(ASCE)CP.1943-5487.0000553
[14]	武立文. 基于SSA-RF模型的岩爆预测方法及应用研究 [D]. 北京: 北方工业大学, 2024. WU L W. Research on rockburst prediction method and application based on SSA-RF model [D]. Beijing: North China University of Technology, 2024.
[15]	刘晓悦, 杨伟, 张雪梅. 基于改进层次法与CRITIC法的多维云模型岩爆预测 [J]. 湖南大学学报(自然科学版), 2021, 48(2): 118–124. doi: 10.16339/j.cnki.hdxbzkb.2021.02.015 LIU X Y, YANG W, ZHANG X M. Rockburst prediction of multi-dimensional cloud model based on improved hierarchical analytic method and critic method [J]. Journal of Hunan University (Natural Sciences), 2021, 48(2): 118–124. doi: 10.16339/j.cnki.hdxbzkb.2021.02.015
[16]	谭文侃, 叶义成, 胡南燕, 等. LOF与改进SMOTE算法组合的强烈岩爆预测 [J]. 岩石力学与工程学报, 2021, 40(6): 1186–1194. doi: 10.13722/j.cnki.jrme.2020.1035 TAN W K, YE Y C, HU N Y, et al. Severe rock burst prediction based on the combination of LOF and improved SMOTE algorithm [J]. Chinese Journal of Rock Mechanics and Engineering, 2021, 40(6): 1186–1194. doi: 10.13722/j.cnki.jrme.2020.1035
[17]	王宇航, 周宗红, 李国才, 等. 基于数据预处理的岩爆等级预测模型及精度优化 [J]. 矿业研究与开发, 2024, 44(11): 101–109. doi: 10.13827/j.cnki.kyyk.2024.11.009 WANG Y H, ZHOU Z H, LI G C, et al. Prediction model and accuracy optimization of rockburst grade based on data preprocessing [J]. Mining Research and Development, 2024, 44(11): 101–109. doi: 10.13827/j.cnki.kyyk.2024.11.009
[18]	翁嘉诚, 周晓杰, 叶蓓蕾, 等. 基于改进麻雀搜索算法的K-means聚类 [J]. 数学的实践与认识, 2024, 54(2): 152–166. WENG J C, ZHOU X J, YE B L, et al. K-means clustering based on improved sparrow search [J]. Mathematics in Practice and Theory, 2024, 54(2): 152–166.
[19]	杜云, 周志奇, 贾科进, 等. 混合多项自适应权重的混沌麻雀搜索算法 [J]. 计算机工程与应用, 2024, 60(7): 70–83. doi: 10.3778/j.issn.1002-8331.2307-0254 DU Y, ZHOU Z Q, JIA K J, et al. Chaotic sparrow search algorithm with mixed multinomial adaptive weights [J]. Computer Engineering and Applications, 2024, 60(7): 70–83. doi: 10.3778/j.issn.1002-8331.2307-0254
[20]	毛清华, 张强. 融合柯西变异和反向学习的改进麻雀算法 [J]. 计算机科学与探索, 2021, 15(6): 1155–1164. doi: 10.3778/j.issn.1673-9418.2010032 MAO Q H, ZHANG Q. Improved sparrow algorithm combining cauchy mutation and opposition-based learning [J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(6): 1155–1164. doi: 10.3778/j.issn.1673-9418.2010032
[21]	张建涛, 刘志祥, 张双侠, 等. 基于WOA-RF的边坡稳定性预测模型 [J]. 高压物理学报, 2024, 38(3): 035301. doi: 10.11858/gywlxb.20230837 ZHANG J T, LIU Z X, ZHANG S X, et al. Slope stability prediction based on WOA-RF hybrid model [J]. Chinese Journal of High Pressure Physics, 2024, 38(3): 035301. doi: 10.11858/gywlxb.20230837