Abstract: Gas is a major threat to coal mine security cause of coal mine threat to security risk,it plays an important role to efficiently and precisely detect gas anomaly for the safety of coal mine production.Traditional gas anomaly detection methods depend solely on gas monitoring data which is often of low reliability due to poor conditions in coal mine and failure of sensors,resulting in false alarm and miss alarm.To solve the above problems,a novel gas anomaly detection algorithm fusing multi-dimensional time series data was proposed based on the time series data from multiple kinds of sensors.The multi-dimensional time series data was sampled with a sliding window,and the Local-sensitive Hashing Isolation Forest was trained with these samples.According to the data to be detected,the anomaly scores and rate were calculated based on the path length of each tree in the forest.When the anomaly rate was greater than the threshold,the forest would be automatically updated.Extensive experiments on real-world Huainan Zhuji coal mine data sets showed that the proposed method achieved higher detection accuracy.

Key words: gas concentration, anomaly detection, sliding windows, locality-sensitive hashing, isolation forest

摘要： 瓦斯是引起煤矿安全隐患的重大因素,高效准确地检测瓦斯异常在煤矿安全生产过程中扮演着重要角色。传统的瓦斯异常检测方法通常仅基于来自瓦斯传感器的单一监测数据,而矿井下的恶劣环境可能造成瓦斯传感器失效,监测数据可信度较低,从而导致误报、漏报等问题。为解决上述问题,基于多种传感器监测数据,提出一种融合煤矿多维时序数据的瓦斯异常检测算法。该方法首先对煤矿中多维时序数据进行滑动窗口采样;然后建立局部敏感哈希孤立森林;最后根据待检测样本在森林中每棵树上的路径长度计算异常得分及异常率,当滑动窗口中的异常率超出指定阈值时,则自动更新森林。通过在真实的淮南朱集煤矿数据集上进行的大量实验,表明了所提方法在提高检测精度上的有效性。

关键词: 瓦斯浓度, 异常检测, 滑动窗口, 局部敏感哈希, 孤立森林