College of Safety Science and Engineering, Liaoning Technical University, Huludao, Liaoning, PR China.
Key Laboratory of Mine Thermal Power Disaster and Prevention of Ministry of Education, Liaoning Technical University, Huludao, Liaoning, PR China.
PLoS One. 2023 Apr 18;18(4):e0284316. doi: 10.1371/journal.pone.0284316. eCollection 2023.
To overcome the false alarm problem that arises for mine wind-velocity sensors due to air-door and mine-car operation, a wind-velocity disturbance identification method based on the wavelet packet transform and gradient lifting decision tree is proposed. In this method, a multi-scale sliding window discretizes continuous wind-velocity monitoring data, the wavelet packet transform extracts the hidden features of discrete data, and a gradient lifting decision tree multi-disturbance classification model is established. Based on the overlap degree rule, the disturbance identification results are merged, modified, combined, and optimized. In accordance with a least absolute shrinkage and selection operator regression, the air-door operation information is further extracted. A similarity experiment is performed to verify the method performance. For the disturbance identification task, the recognition accuracy, accuracy, and recall of the proposed method are 94.58%, 95.70% and 92.99%, respectively, and for the task involving further extraction of disturbance information related to air-door operation, those values are 72.36%, 73.08%, and 71.02%, respectively. This algorithm gives a new recognition method for abnormal time series data.
为了解决风门和矿车作业引起的矿山风速传感器误报问题，提出了一种基于小波包变换和梯度提升决策树的风速干扰识别方法。该方法采用多尺度滑动窗口对连续风速监测数据进行离散化，利用小波包变换提取离散数据的隐含特征，建立梯度提升决策树多干扰分类模型。根据重叠度规则对干扰识别结果进行合并、修正、组合和优化，进一步采用最小绝对值收缩和选择算子回归提取风门作业信息。通过相似性实验验证了该方法的性能。在干扰识别任务中，该方法的识别准确率、精度和召回率分别为 94.58%、95.70%和 92.99%，而在涉及进一步提取与风门作业相关的干扰信息的任务中，这些值分别为 72.36%、73.08%和 71.02%。该算法为异常时间序列数据提供了一种新的识别方法。
