摘要
煤矿智能管控面临动态响应滞后和多源数据割裂的挑战,针对传统模型难以捕捉井下瞬态异常和协同分析多模态数据的问题。文章提出基于Transformer的多任务自适应架构(MTA-Transformer),通过跨模态特征融合与共享编码器,统一建模设备振动、瓦斯浓度等数据,实现开采环境的多尺度动态感知,解决对开采环境的动态监控与风险超前预警问题。实验表明,在轴承故障检测任务中,该模型准确率达93.5%,误报率(FAR)为2.0%,响应时间在5 ms内,较传统模型有较大提升;瓦斯浓度预测NRMSE为7.83%,预测区间覆盖概率(PICP)达91.7%,超前预警时效可达6 h。MTA-Transformer为矿山智能化建设提供了可落地的模式。
The intelligent control of coal mines is facing the challenges of dynamic response lag and multi-source data fragmentation,while traditional models are incapable to capture transient anomalies underground and collaboratively analyze multimodal data.We propose a Multi-Task Adaptive Transformer Architecture(MTA-Transformer).Through cross-modal feature fusion and a shared encoder,it unifies the modeling of data such as equipment vibration and gas concentration,achieving multi-scale dynamic perception of the mining environment.This solves the problems of dynamic environmental monitoring and early risk warning.Experiments show that for bearing fault detection,the model achieves an accuracy of 93.5%,a False Alarm Rate(FAR)of 2.0%,and a response time within 5 ms,representing significant improvement over traditional models.For gas concentration prediction,it achieves a Normalized Root Mean Square Error(NRMSE)of 7.83%,a Prediction Interval Coverage Probability(PICP)of 91.7%,and enables early warnings up to 6 hours in advance.The MTA-Transformer provides a practical technical paradigm for the intelligent construction of mines.
作者
牛云鹏
索智文
王惠伟
屈波
周超逸
张丽芳
NIU Yunpeng;SUO Zhiwen;WANG Huiwei;QU Bo;ZHOU Chaoyi;ZHANG Lifang(CHN Shendong Coal Intelligent Technology Center,Yulin 719315,China;China Academy of Safety Science and Technology,Beijing 100012,China)
出处
《煤炭工程》
北大核心
2026年第1期35-42,共8页
Coal Engineering
基金
中国神华能源股份有限公司神东煤炭分公司科技创新项目(E210100270)。
关键词
煤矿智能管控
TRANSFORMER
多任务协同
故障诊断
瓦斯浓度预测
intelligent coal mine management and control
Transformer
multi-task cooperation
fault diagnosis
gas concentration prediction
