[1]王 强,李 军,李长旭,等.基于骨架序列的发电厂作业人员异常行为识别研究[J].机械与电子,2026,44(07):39-45.
　WANG Qiang,LI Jun,LI Changxu,et al.Research on Abnormal Behavior Recognition of Power Plant Operators Based on Skeleton Sequence[J].Machinery & Electronics,2026,44(07):39-45.
点击复制

基于骨架序列的发电厂作业人员异常行为识别研究()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年07期
页码:
39-45
栏目:
智能检测
出版日期:
2026-07-25

文章信息/Info

Title:
Research on Abnormal Behavior Recognition of Power Plant Operators Based on Skeleton Sequence
文章编号:
1001-2257(2026)07-0039-07
作者:
王 强; 李 军; 李长旭; 万培文; 马亚聪
国家能源集团宁夏电力有限公司,宁夏 银川 750002
Author(s):
WANG Qiang; LI Jun; LI Changxu; WAN Peiwen; MA Yacong
(State Energy Group Ningxia Electric Power Co.,Ltd.,Yinchuan 750002,China)
关键词:
计算机视觉; 行为识别; 电磁鲁棒时空图; 张量鲁棒主成分分析; 双支路异构特征融合
Keywords:
computer vision; recognition; behavior electromagnetic-robust spatio-temporal graph; tensor robust principal component analysis; dual-branch heterogeneous feature fusion
分类号:
TP391.4
文献标志码:
A
摘要:
针对发电厂强电磁、粉尘与抖动耦合致骨架识别拓扑失配、数据退化及罕见异常召回不足问题,提出电磁鲁棒时空异构协同网络(STHS Net)。首先建立电磁量化时空图(ERSTG),将电磁强度嵌入动态拓扑剪枝并引入帧间记忆,抑制虚假连接;再提出耦合噪声分离质量再生策略(CNSQR),以张量鲁棒主成分分解联合骨骼长度守恒正则,实现低秩补全;继而利用双支路异构特征融合(DBHFF)编码势能突变与异常激活门控深度特征,构建物理保持、深度异常和全局上下文3路协同网络,配合梯度平衡重加权提升罕见异常召回。实验表明,相比GCN-RNN、STTN等方法,STHS-Net的mAP提高2.1~10.4百分点,RCR提高3.3~12.7百分点,单帧耗时8.7 ms、帧率115 帧/s、模型大小28.6 MB,为强扰动工业场景提供了可部署方案。
Abstract:
To address the challenges of topological mismatch in skeleton recognition,data degradation, and insufficient recall of rare anomalies,caused by the coupling of strong electromagnetic interference, dust,and vibration in power plants,this paper proposes an Electromagnetic-Robust Spatio-TemporalHeterogeneous Synergy Network (STHS-Net).First,an electromagnetic-quantified spatiotemporal graph (ERSTG) is constructed by embedding measured electromagnetic intensity into dynamic topology and pruning with inter-frame memory updating to suppress spurious connections.Then,a coupled noise separation and quality regeneration strategy (CNSQR) is proposed,employing tensor robust principal component analysis combined with bone length preservation regularization to achieve low-rank completion. Subsequently,a dual-branch heterogeneous feature fusion (DBHFF) module encodes potential-energy mutation and anomaly activated gated deep features,forming a tripartite collaborative network comprising physically preserved,deep anomaly,and global context pathways.A gradient-balanced re-weighting mechanism is further adopted to enhance the recall of rare abnormal behaviors.Experimental results show that,compared with methods such as GCN-RNN and STTN,STHS-Net improves the mean average precision (mAP) by 2.1—10.4 percentage points and rare-class recall (RCR) by 3.3—12.7 percentage points,while achieving a per-frame inference time of 8.7 ms,a frame rate of 115 FPS,and a model size of 28.6 MB,providing a deployable solution for strong-disturbance industrial scenarios.

参考文献/References:

[1] 于作华,陶智宇,李晓辉,等.水电站智能巡检系统监测运行与时序特征分析[J].大电机技术,2024(5):117-127.

[2] 许素睿.燃煤电厂生产性粉尘健康风险评价[J].中国安全科学学报,2020,30(2):152-157.
[3] 董佳康,何涛.基于自适应图拓扑细化与跨序列对比学习的骨架动作识别优化[J].计算机系统应用,2025,34(11):107-114.
[4] 余向前,喻见,王婷婷,等.基于挤压激励卷积神经网络的配电网拓扑辨识[J].电网与清洁能源,2025,41(8):83-89.
[5] 董贇,张希翔,艾徐华,等.基于改进AlphaPose的人体特征行为识别方法及应用[J].武汉大学学报(工学版),2025,58(11):1767-1775.
[6] 祁升龙,何玉鹏,张仁和,等.基于概率化时序区间的行为识别及其在电力巡检中的应用[J].计算机应用与软件,2025,42(3):61-67.
[7] 党伟超,张泽杰,白尚旺,等.基于改进双流法的井下配电室巡检行为识别[J].工矿自动化,2020,46(4):75-80.
[8] 吴志泽,万龙,洪芳华,等.融合骨架大核算子和全局上下文信息的图卷积网络[J].中国图象图形学报,2025,30(11):3604-3616.
[9] 孟祥璞,李硕,苑明哲,等.基于人体骨架的动作识别:综述与展望[J].信息与控制,2025,54(1):1-27.
[10] 赵星,白建豪,傅迎华.基于张量鲁棒主成分分析的视网膜眼底图像序列变化检测[J].信息与控制,2023,52(1):115-128.
[11] Yang Yijie,Zhang Jinlu,Zhang Jiaxu,et al.Expressive keypoints for skeleton-based action recognition via progressive skeleton evolution[J].IEEE Transactions on Image Processing,2025,34:7585-7599.
?[12] Zhou Haiyun,Xiang Xuezhe,Qiu Yujian,et al.Graph convolutional network with STC attention and adaptive normalization for skeleton-based action recognition[J].The imaging science journal,2023,71(7):636-646.
[13] Kilic U,Karadag O O,Ozyer G T.AGMS-GCN:attention-guided multi-scale graph convolutional networks for skeleton-based action recognition[J]. Knowledge-Based Systems,2025,311:113045.
[14] Guendoul O,Zobi M,Abdelali H A,et al.Capturing spatio-temporal patterns of falls individuals using efficient graph convolutional network model[J].Applied Intelligence,2025,55(11):1-16.
[15] Lovanshi M,Tiwari V,Ingle R,et al.Fusion of temporal transformer and spatial graph convolutional network for 3-D skeleton-parts-based human motion prediction[J].IEEE Transactions on Human-Machine Systems,2024,54(6):788-797.

相似文献/References:

[1]陈 红,胡恩才,白静静.交叉路口信号灯智能管理系统设计[J].机械与电子,2019,(11):63.
　,Design of Intelligent Management System for Intersection Signal Lights[J].Machinery & Electronics,2019,(07):63.
[2]鲍克鹏,许 涛,贾雪磊,等.基于改进双边分割网络的视觉测振方法[J].机械与电子,2025,(03):29.
　BAO Kepeng,XU Tao,JIA Xuelei,et al.Visual Vibration Measurement Method Based on an Improved Bilateral Segmentation Network[J].Machinery & Electronics,2025,(07):29.

备注/Memo

备注/Memo:
收稿日期:2026-02-07
基金项目:国家重点研发计划项目(2022YFB2403200)
作者简介:王 强 (1979-),男,北京人,硕士,高级工程师,研究方向为电力市场运营。
更新日期/Last Update: 2026-08-27