【摘要】 传统的聚类异常数据检测算法在处理高维度、大数据量且异常值分布杂乱的机电设备环境参数时，存在聚类效果差和检测效率低的问题。为此，在原有异常检测算法的基础上提出一种基于先验聚类的机电设备环境参数异常检测算法。该算法改用历史数据构建先验聚类，确保聚类构建不会受太多异常环境参数所影响；在选取聚类中心时引入密集度的概念，以确保聚类中心的可靠性，并在选取聚类中心过程中去除已选聚类中心周围的数据点，防止选取的聚类中心集中在某一区域，以此提升聚类效果。进行异常检测时，依次将待检测数据放入先验聚类中进行匹配，一旦测试数据无法匹配任何一个已知聚类，则将其标记为异常数据。实验结果表明：所提算法在机电设备环境参数的异常检测方面具有检测率高、误报率低的特点，在2 000例数据异常检测中，其检测准确率达到了97.5%，优于DBSCAN算法的97%以及基础K-means算法的86%；同时，误检率低至0.010 6，优于DBSCAN算法的0.023 9和基础K-means算法的0.022 8。改进后的模型较基础K-means算法和DBSCAN算法在机电设备环境参数异常检测中检测效果更佳，在机电设备环境异常数据检测上具有良好的性能。更多还原

【Abstract】 The traditional clustering anomaly data detection algorithm has the problem of poor clustering effect and low detection efficiency when dealing with the environmental parameters of electromechanical equipment with high dimension, large data amount and chaotic distribution of outliers. Therefore, on the basis of the traditional anomaly detection algorithm, a prior clustering based environmental parameter anomaly detection algorithm of electromechanical equipment is proposed. In this algorithm, the historical data is used to construct prior clustering to ensure that the cluster construction cannot be affected by too many abnormal environmental parameters. The concept of density is introduced to ensure the reliability of cluster centers when selecting cluster centers, and the data points around the selected cluster centers are removed in the process of selecting cluster centers to prevent the selected cluster centers from being concentrated in a certain area, so as to improve the clustering effect. In the process of anomaly detection, the data to be detected are put into the prior clustering for matching. Once the testing data cannot match any of the known clusters, it is marked as abnormal data. The experimental results show that the proposed algorithm has the characteristics of high detection rate and low false positive rate in the abnormal detection of electromechanical equipment environmental parameters. In the abnormal detection of 2 000 cases of data, the detection accuracy rate can reach 97.5%, which is better than 97% of DBSCAN algorithm and 86% of basic K-means algorithm. Its false detection rate is as low as 0.010 6, which is better than 0.023 9 of DBSCAN algorithm and 0.022 8 of basic K-means algorithm. In comparison with basic Kmeans algorithm and DBSCAN algorithm, the improved model has better detection effect in the environmental parameters anomaly detection of electromechanical equipment, and has good performance in the detection of environmental abnormal data of electromechanical equipment.更多还原