基于M估计量及标准四分位间距的安全监测数据异常识别的改进方法

An Improved Method of Anomaly Recognition of Dam Safety Monitoring Data Based on M-Estimator and Standard Quartile Range

水电工程安全监测数据的异常识别是科学评价大坝安全性态的前提。为了解决传统的3σ准则运用于“台阶型”、“震荡型”监测数据序列异常在线识别时极易出现的漏判问题,引入Andrews M估计量和标准四分位间距替代传统3σ准则中的总体位置参数和总体尺度参数,构建了新的判别准则。工程实践及敏感性分析表明:该方法能有效消除“台阶型”和“震荡型”离群点对识别结果带来的不利影响;抗离群点影响的比例提升到25%,异常识别的准确性及可靠性提升明显。研究成果可为水电工程安全状况和运行性态的评价提供参考。

Anomaly recognition of safety monitoring data of hydropower station is a prerequisite for scientific evaluation of dam safety.Traditional 3σcriterion is prone to cause miss judgment when applied to the online anomaly identification of“step type”and“oscillating type”monitoring data series.In view of this,we established an improved criterion by replacing the general position parameter and general scale parameter in the 3σcriteria with Andrews M-estimator and standard quartile range.Engineering practice and sensitivity analysis prove that the method could effectively eliminate the adverse effects of anomalies on the recognition results.The proportion of anti-anomaly amounts 25%,and the accuracy and reliability of anomaly recognition are improved obviously.