煤矿生产安全对国家能源供应稳定至关重要，而矿工的隐患识别能力不足是安全隐患排查治理的关键挑战之一。现场隐患识别本质上是视觉搜索任务，经验丰富者具备更高效的隐患识别视觉搜索策略，但此类隐性知识难以通过传统培训项目中的文本或口述的方式有效传授，亟需一种将内隐知识外显化的安全培训措施。基于此，本研究在模拟井下场景中开展隐患识别眼动实验，采集到68名矿工的有效眼动数据和行为数据，采用NASA-TXL量表评估任务主观认知负荷。运用K-means聚类方法将样本分为高、中、低绩效三组，深入分析不同绩效组间各项眼动指标的差异，揭示高效的隐患识别视觉搜索策略。引入以个性化绩效反馈、眼动可视化反馈和元认知干预为举措的个性化训练干预措施，对无经验矿工进行培训。通过开展眼动实验采集到31名干预训练前与21名干预训练后被试的有效眼动数据，验证措施对矿工隐患识别能力的提升效果。结合鲸鱼优化算法（Whale Optimization Algorithm，WOA）与支持向量机（Support Vector Machine，SVM），构建以眼动数据驱动的矿工隐患识别结果预测模型。研究结果表明：

（1）眼动数据能够将高绩效矿工群体内隐的高效隐患识别视觉搜索策略外显化。实验样本内矿工的经验水平与隐患识别绩效显著正相关，知识水平对绩效无显著影响；多项眼动指标表明高绩效组矿工采用了“全局搜索-关键区域聚焦-回访确认”的系统化策略，实现视觉注意力广覆盖且高效的隐患识别；眼动热点图与眼动轨迹图直观反映矿工采取的隐患识别视觉搜索策略，有效将相关隐性知识外显化。

（2）引入的个性化干预措施显著提升无经验矿工的隐患识别能力。干预训练后的隐患识别正确率平均值提高37.57%，被试均达到高绩效水平，且主观认知负荷较干预训练前下降26.34%；对隐患区域的注视时间和次数显著增加，视觉注意力分配更具目标导向，视觉搜索行为更趋系统化；干预训练后多项眼动指标与高绩效矿工组无显著差异，采用视觉搜索策略与高绩效矿工组趋同，干预措施有效提升矿工隐患识别能力。

（3）构建的WOA-SVM隐患识别结果预测模型具备较高预测准确性。通过特征筛选得到9项关键眼动特征，其中访问次数和平均注视时长对于结果预测贡献最大；采用基于编辑最近邻规则欠采样与合成少数类过采样（Edited Nearest Neighbor-Synthetic Minority Oversampling Technique，ENN-SMOTE）算法将数据集不平衡率降低至1.02，通过WOA算法优化SVM模型的超参数，对隐患识别结果预测准确率达91.14%，精确度达91.45%，表现出良好的预测性能。

研究结果进一步深化了煤矿隐患识别视觉认知机制的理解，通过眼动追踪技术实现隐患识别隐性知识的外显化，制定的个性化干预措施突破了传统安全培训单向知识灌输模式，推动隐患识别能力提升由知识经验积累型向基于视觉认知机制的精准干预型转变，为煤矿安全培训模式的转型升级提供了数据支持和实践参考。

Coal mine production safety is crucial for ensuring the stability of national energy supply, while miners' inadequate hazard identification abilities remain a key challenge in safety hazard detection and remediation. Hazard identification in the field is essentially a visual search task, and experienced workers possess more efficient visual search strategies for hazard identification. However, this implicit knowledge is difficult to effectively convey through traditional training programs based on text or oral instructions, necessitating a safety training approach that externalizes tacit knowledge. Based on this, this study conducted a hazard identification eye-tracking experiment in a simulated underground mining scenario, collecting valid eye-tracking and behavioral data from 68 miners, and using the NASA-TLX scale to assess the subjective cognitive load of the tasks. The samples were divided into high, medium, and low-performance groups using the K-means clustering method. Differences in eye-tracking indicators between the groups were analyzed in-depth to reveal efficient hazard identification visual search strategies. A personalized training intervention involving performance feedback, eye-tracking visualization feedback, and metacognitive interventions was introduced for inexperienced miners. Valid eye-tracking data were collected from 31 participants before training and 21 after training, and comparative analysis was conducted to verify the effectiveness of the intervention in improving miners' hazard identification ability. A hazard identification result prediction model driven by eye-tracking data was built using Whale Optimization Algorithm (WOA) and Support Vector Machine (SVM).

The results indicate that: (1) Eye-tracking data can externalize the efficient hazard identification visual search strategies of high-performance miners. There is a significant positive correlation between miners' experience level and hazard identification performance, while knowledge level does not significantly impact performance. Multiple eye-tracking indicators show that high-performance miners employ a systematic strategy of "global search - key area focus - revisit confirmation," achieving broad and efficient hazard identification with visual attention. Eye-tracking heatmaps and trajectory maps can visually reflect the visual search strategies adopted by miners, making them suitable as materials for personalized training interventions. (2) The introduced personalized intervention measures significantly improved the hazard identification ability of inexperienced miners. The average hazard identification accuracy after intervention increased by 37.57%, with all participants reaching high-performance levels, and subjective cognitive load decreased by 26.34% compared to pre-intervention. The time and frequency of fixation on hazard areas significantly increased, with more targeted visual attention distribution, and visual search behavior became more systematic. After the intervention, several eye-tracking indicators showed no significant differences from the high-performance group, and the visual search strategy for hazard identification converged with that of high-performance miners. (3) The WOA-SVM hazard identification result prediction model demonstrates high prediction accuracy. Nine key eye-tracking features were identified through feature selection, with visit frequency and average fixation duration contributing the most to result prediction. By using the ENN-SMOTE algorithm, the dataset's imbalance rate was reduced to 1.02, and the SVM model's hyperparameters were optimized using WOA. The hazard identification result prediction accuracy reached 91.14%, with a precision of 91.45%, showing excellent predictive performance.

The research findings further deepen the understanding of the visual cognitive mechanisms underlying hazard identification in coal mines. By utilizing eye-tracking technology, the study externalizes the tacit knowledge involved in hazard recognition. The personalized intervention measures developed in this research break through the traditional one-way knowledge dissemination model of safety training, promoting a shift from experience-based knowledge accumulation to precision intervention based on visual cognition mechanisms. This provides data support and practical reference for the transformation and upgrading of coal mine safety training models.

[1] 吴奎斌, 吴昊, 李瑞华. “双碳”战略下煤矿减碳实施路径研究[J]. 中国煤炭, 2024, 50(5): 13-17.

[2] 杜伟,文腾. 《"十四五"现代能源体系规划》等多项政策出台布局中国新型能源体系[J]. 国际石油经济,2023,31(1):9-10.

[3] SU G, HU E. Research on coal mine safety risk evolution and key hidden dangers under the perspective of complex network[J]. Scientific Reports, 2024, 14(1): 20624

[4] 《关于防范遏制矿山领域重特大生产安全事故的硬措施》印发[J].中国煤炭工业,2024,(02):40.

[5] EITER B M, BELLANCA J L. Identify the Influence of Risk Attitude, Work Experience, and Safety Training on Hazard Recognition in Mining[J]. Mining, Metallurgy & Exploration, 2020, 37(6): 1931-1939

[6] EITER B M, BELLANCA J L, HELFRICH W, et al. Recognizing Mine Site Hazards: Identifying Differences in Hazard Recognition Ability for Experienced and New Mineworkers[C]//Advances in Human Factors in Simulation and Modeling. Springer, Cham, 2018: 104-115.

[7] BARRETT E A, KOWALSKI-TRAKOFLER K M. Effective hazard recognition training using a latent-image, three-dimensional slide simulation exercise[J]. 1995.

[8] JEELANI I, ALBERT A, HAN K, et al. Are Visual Search Patterns Predictive of Hazard Recognition Performance? Empirical Investigation Using Eye-Tracking Technology[J]. Journal of Construction Engineering and Management, 2019, 145(1): 04018115.

[9] NAMIAN M, ALBERT A, ZULUAGA C M, et al. Role of Safety Training: Impact on Hazard Recognition and Safety Risk Perception[J]. Journal of Construction Engineering and Management, 2016, 142(12): 04016073.

[10] Albert A, Jeelani I, Han K. Developing hazard recognition skill among the next-generation of construction professionals[J]. Construction management and economics, 2020, 38(11): 1024-1039.

[11] JEELANI I, ALBERT A, AZEVEDO R, et al. Development and Testing of a Personalized Hazard-Recognition Training Intervention[J]. Journal of Construction Engineering and Management, 2017, 143(5): 04016120.

[12] 徐晴雯. 工程安全隐患视觉搜索策略研究[D]. 北京:清华大学, 2021.

[13] DZENG R J, LIN C T, FANG Y C. Using eye-tracker to compare search patterns between experienced and novice workers for site hazard identification[J]. Safety Science, 2016, 82: 56-67.

[14] ZHOU Z, GOH Y M, LI Q. Overview and analysis of safety management studies in the construction industry[J]. Safety Science, 2015, 72: 337-350.

[15] 程瑞. 施工安全隐患识别视觉认知策略研究[D]. 北京: 清华大学, 2022.

[16] GEGENFURTNER A, LEHTINEN E, SÄLJÖ R. Expertise differences in the comprehension of visualizations: A meta-analysis of eye-tracking research in professional domains[J]. Educational psychology review, 2011, 23: 523-552.

[17] Shen S H. IDA: a cognitive model of nuclear power plant operator response during abnormal conditions[M]. University of Maryland, College Park, 1994.

[18] CHANG Y H J, MOSLEH A. Cognitive modeling and dynamic probabilistic simulation of operating crew response to complex system accidents. Part 2: IDAC performance influencing factors model[J]. Reliability Engineering & System Safety, 2007, 92(8): 1014-1040.

[19] KOWALSKI-TRAKOFLER K M, BARRETT E A. The concept of degraded images applied to hazard recognition training in mining for reduction of lost-time injuries[J]. Journal of Safety Research, 2003, 34(5): 515-525.

[20] FANG D, ZHAO C, ZHANG M. A Cognitive Model of Construction Workers’ Unsafe Behaviors[J]. Journal of Construction Engineering and Management, 2016, 142(9): 04016039.

[21] ZHENG X, WANG Y, CHEN Y, et al. Influence of Safety Experience and Environmental Conditions on Site Hazard Identification Performance[J]. Buildings, 2023, 13(1): 251.

[22] HIGGINS E, LEINENGER M, RAYNER K. Eye movements when viewing advertisements[J]. Frontiers in psychology, 2014, 5: 210.

[23] WICKENS C D, HELTON W S, HOLLANDS J G, et al. Engineering psychology and human performance[M]. Routledge, 2021..

[24] PANNASCH S, HELMERT J R, ROTH K, et al. Visual Fixation Durations and Saccade Amplitudes: Shifting Relationship in a Variety of Conditions[J]. Journal of Eye Movement Research, 2008, 2(2).

[25] THORNTON T L, GILDEN D L. Parallel and serial processes in visual search[J]. Psychological Review, 2007, 114(1): 71-103.

[26] DRURY C G. Inspection of Sheet Materials — Model and Data[J]. Human Factors: The Journal of the Human Factors and Ergonomics Society, 1975, 17(3): 257-265.

[27] 沈文意,于战宇,冷英.语义大小与刺激数量在非对称性视觉搜索中的作用[J].心理研究,2024,17(1):26-33.

[28] 韩振华, 曹立人. 平行搜索还是系列搜索——视觉搜索机制研究的理论分析[J]. 西北师大学报（社会科学版）, 2009, 46(5): 129-132.

[29] ZHANG Q, LIANG M, CHAN A P C, et al. Visual attention and cognitive process in construction hazard recognition: Study of fixation-related potential[J]. Automation in Construction, 2023, 148: 104756.

[30] LIU M, LIAO P C, WANG X Y, et al. Influence of semantic cues on hazard-inspection performance: a case in construction safety[J]. International Journal of Occupational Safety and Ergonomics, 2021, 27(1): 14-2.

[31] OUYANG Y, LUO X. Differences between inexperienced and experienced safety supervisors in identifying construction hazards: Seeking insights for training the inexperienced[J]. Advanced Engineering Informatics, 2022, 52: 101602.

[32] LI S, JIANG Y, SUN C, et al. An Investigation on the Influence of Operation Experience on Virtual Hazard Perception Using Wearable Eye Tracking Technology[J]. Sensors, 2022, 22(14): 5115.

[33] 郑霞忠, 石博元, 陈云, 等. 融合VR-眼动的施工现场隐患识别视觉注意与搜索特征研究[J]. 安全与环境学报, 2024, 24(3): 1087-1095.

[34] ZHAO S Q, XIANG C lang, SU L, et al. Study on the Eye Movement Characteristics of Fire Hazard Identification in University Laboratories[J]. Procedia Engineering, 2018, 211: 433-440.

[35] HASANZADEH S, ESMAEILI B, DODD M D. Measuring Construction Workers’ Real-Time Situation Awareness Using Mobile Eye-Tracking[J]. 2016: 2894-2904.

[36] LIAO P C, SUN X, LIU M, et al. Influence of visual clutter on the effect of navigated safety inspection: a case study on elevator installation[J]. International Journal of Occupational Safety and Ergonomics, 2019, 25(4): 495-509.

[37] XU Q, CHONG H Y, LIAO P chao. Exploring eye-tracking searching strategies for construction hazard recognition in a laboratory scene[J]. Safety Science, 2019, 120: 824-832.

[38] 倪国栋, 方亚琦, 张琦, 等. 风险倾向对新生代建筑工人危险感知的影响机制[J]. 中国安全科学学报, 2023, 33(5): 221-229.

[39] DU X, MA J, CHANG R. The interactive effect of vehicle signals and sensation-seeking on driver hazard perception[J]. Transportation Research Part F: Traffic Psychology and Behaviour, 2020, 73: 174-187.

[40] HASLAM R A, HIDE S A, GIBB A G F, et al. Contributing factors in construction accidents[J]. Applied Ergonomics, 2005, 36(4): 401-415.

[41] 浦仕跃. 煤矿安全培训与技能提升研究[J]. 内蒙古煤炭经济, 2024(11): 88-90.

[42] HAN Y, YANG J, DIAO Y, et al. Process and Outcome-based Evaluation between Virtual Reality-driven and Traditional Construction Safety Training[J]. Advanced Engineering Informatics, 2022, 52: 101634.

[43] LIU P, ZHANG C, ARDITI D, et al. Using eye-tracking technology to assess the effect of daily safety training on hazard recognition skills[J]. IEEE Transactions on Engineering Management, 2024, 71: 8548-8561.

[44] JEELANI I, ALBERT A, GAMBATESE J A. Why Do Construction Hazards Remain Unrecognized at the Work Interface?[J]. Journal of Construction Engineering and Management, 2017, 143(5): 04016128.

[45] FRAME M E, WARREN R, MARESCA A M. Scanpath comparisons for complex visual search in a naturalistic environment[J]. Behavior Research Methods, 2019, 51(3): 1454-1470.

[46] LOOSEMORE M, MALOUF N. Safety training and positive safety attitude formation in the Australian construction industry[J]. Safety Science, 2019, 113: 233-243.

[47] TIEN T, PUCHER P H, SODERGREN M H, et al. Eye tracking for skills assessment and training: a systematic review[J]. Journal of Surgical Research, 2014, 191(1): 169-178.

[48] TUNGA Y, CAGILTAY K. Looking through the model’s eye: A systematic review of eye movement modeling example studies[J]. Education and Information Technologies, 2023, 28(8): 9607-9633.

[49] 付汉良,谭玉冰,夏中境,等.专家危险识别轨迹对建筑工人安全教育的影响——来自眼动实验的证据[J].清华大学学报(自然科学版),2024,64(02):205-213.

[50] LUKE M M, ALAVOSIUS M. Adherence with universal precautions after immediate, personalized performance feedback.[J]. Journal of Applied Behavior Analysis, 2011, 44(4): 967-971.

[51] JEELANI I, HAN K, ALBERT A. Automating and scaling personalized safety training using eye-tracking data[J]. Automation in Construction, 2018, 93: 63-77.

[52] 梁逸飞. 基于眼动追踪技术的建筑施工危险认知研究[D]. 广东: 深圳大学, 2023.

[53] 石祝, 尚俊杰. 视频游戏对空间能力的影响与作用机制[J]. 中国电化教育, 2024(5): 32-44, 113.

[54] DRURY C G. Integrating Human Factors Models into Statistical Quality Control[J]. Human Factors, 1978, 20(5): 561-572.

[55] DUNCAN J, HUMPHREYS G. Beyond the search surface: Visual search and attentional engagement[J]. Journal of Experimental Psychology: Human Perception and Performance, 1992, 18(2): 578-588.

[56] 陈学强, 黄黎清, 李明珠. 城际物流APP界面导航设计的视觉搜索绩效研究[J]. 包装工程, 2021, 42(8): 198-204.

[57] 廖斌, 冯海芹, 吴雪琴. 气质与VDT持续搜索策略及人因可靠性的关系[J]. 安全与环境学报, 2021, 21(4): 1615-1621.

[58] 于瑞峰, 孔哲昕. 视觉搜索策略研究综述[J]. 人类工效学, 2012, 18(2): 93-95.

[59] 吴禧芊, 张西磊, 蒋毅, 等. 选择性注意对无意识加工的调节作用及潜在机制[J]. 生物化学与生物物理进展, 2024, 51(9): 2016-2027.

[60] 邓宏泽, 孔媛媛, 徐晟. 不同认知负荷下施工现场隐患识别的视觉行为研究[J]. 中国安全科学学报, 2025, 35(1): 40-49.

[61] MULEY S, WANG C, AGHAZADEH F. Personality traits affecting construction Worker's near-miss recognition performance: Analysis based on eye tracking[J]. International Journal of Industrial Ergonomics, 2024, 102: 103606.

[62] 谢斌红, 马非, 潘理虎, 等. 煤矿安全隐患信息自动分类方法[J]. 工矿自动化, 2018, 44(10): 10-14.

[63] 李思宇. 风险感知能力影响的眼动实验设计及研究[D]. 黑龙江:哈尔滨理工大学, 2023.

[64] 姜婷婷, 吴茜, 徐亚苹, 等. 眼动追踪技术在国外信息行为研究中的应用[J]. 情报学报, 2020, 39(2): 217-230.

[65] JEELANI I, HAN K, ALBERT A. Automating and scaling personalized safety training using eye-tracking data[J]. Automation in Construction, 2018, 93: 63-77.

[66] PAPESH M H, HOUT M C, GUEVARA PINTO J D, et al. Eye movements reflect expertise development in hybrid search[J]. Cognitive research: principles and implications, 2021, 6: 1-20.

[67] 曾娟, 王昊, 许博, 等. 基于强弱感知设计的驾驶员危险感知状态识别模型研究[J]. 汽车工程, 2024, 46(6): 995-1005.

[68] HART S G. NASA-task load index (NASA-TLX); 20 years later[C]//Proceedings of the human factors and ergonomics society annual meeting. Sage CA: Los Angeles, CA: Sage publications, 2006, 50(9): 904-908.

[69] HASANZADEH S, ESMAEILI B, DODD M D. Measuring the Impacts of Safety Knowledge on Construction Workers’ Attentional Allocation and Hazard Detection Using Remote Eye-Tracking Technology[J]. Journal of Management in Engineering, 2017, 33(5): 04017024.

[70] HASANZADEH S, ESMAEILI B, DODD M D. Impact of Construction Workers’ Hazard Identification Skills on Their Visual Attention[J]. Journal of Construction Engineering and Management, 2017, 143(10): 04017070.

[71] TREISMAN A M, GELADE G. A feature-integration theory of attention[J]. Cognitive Psychology, 1980, 12(1): 97-136.

[72] BONETTI L V, HASSAN S A, LAU S T, et al. Oxyhemoglobin changes in the prefrontal cortex in response to cognitive tasks: a systematic review[J]. International Journal of Neuroscience, 2019, 129(2): 195-203.

[73] FLAVELL J H. Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry.[J]. American Psychologist, 1979, 34(10): 906-911.

[74] 沈学利, 杨莹, 秦鑫宇, 等. 基于残差神经网络的风机叶片结冰故障诊断[J]. 噪声与振动控制, 2022, 42(1): 79-87.

[75] CHAUHAN V K, DAHIYA K, SHARMA A. Problem formulations and solvers in linear SVM: a review[J]. Artificial Intelligence Review, 2019, 52(2): 803-855.

[76] 程子鉴, 陈星明, 安英东, 等. 基于WOA-SVM模型的边坡安全系数预测[J]. 有色金属（矿山部分）, 2025, 77(1): 159-165.

[77] 丁茜, 赵晓东, 吴鑫俊, 等. 基于RBF核的多分类SVM滑塌易发性评价模型[J]. 中国安全科学学报, 2022, 32(3): 194-200.

[78] 张晓鹏,秦亮曦.一种不平衡数据多策略处理及组合分类算法[J].计算机应用与软件, 2024, 41(4): 242-250+305.

[79] 刘奇,梁智昊,訾建潇.SMOGN过采样下导水裂隙带高度的MPSO-BP预测模型[J].煤田地质与勘探,2024,52(11):72-85.

[80] GUO J, WU H, CHEN X, et al. Adaptive SV-Borderline SMOTE-SVM algorithm for imbalanced data classification[J]. Applied Soft Computing, 2024, 150: 110986.

[81] 张爱民, 于化龙. 基于OCkNN+ENN的过采样算法研究[J]. 计算机与数字工程, 2024, 52(5): 1275-1281.

[82] MIRJALILI S, LEWIS A. The Whale Optimization Algorithm[J]. Advances in Engineering Software, 2016, 95: 51-67.