基于影响因子分解法的大坝监测数据异常检测算法

Detection method for dam abnormal monitoring data based on influcing factor decomposition

如何快速检测出大坝安全监测系统内的异常数据(例如粗差和告警值)对于大坝安全运行具有极其重要的意义,但传统方法容易漏检较小数值异常而对后续建模产生不利影响。提出了一种基于影响因子分解的异常值检测方法,通过快速小波变换及离散傅里叶变换提取监测序列中的显著趋势与周期,剥离环境因子的影响,构建余项序列,并结合小概率事件思想准确判定余项序列中保留的异常值,从而精确检测出监测序列中较小数值异常。实例验证结果表明:此方法具有较好的实用性与稳定性,各类监测序列中异常检测准确率均达98%以上,查准率与查全率均值分别为93%与92%,与传统检测方法相比,检测精确程度及泛化能力明显提升。

Rapid detection of abnormal data in the dam safety monitoring system(such as coarse difference and alarm value)is significant for the safe operation of dams.But traditional methods are prone to miss the detection of small numerical anomalies thus adversely affect subsequent modeling.In this paper,an abnormal monitoring data detection method based on influcing factor decomposition is proposed.It can extract the significant trends and periods in the monitoring sequence by the rapid wavelet transform and the discrete fourier transform,strip away the influence of environmental factors to construct the remainder sequence,and further accurately determine the abnormal monitoring data retained in the remainder sequence in combination with the idea of small probability events.Finally it accurately detects the abnormal data in the monitoring sequence.The numerical results showed that the proposed method has good practicality and stability,the accuracy rate of abnormal detection of various monitoring sequences is more than 98%,and the average values of precision and recall rate are 93%and 92%respectively,showing certainly improved accuracy and generalization ability compared with the traditional detection methods.