【作者】 罗诗怡； 谷艳昌； 陆强； 吴云星； 黄之源；

【Author】 Shiyi Luo;Yanchang Gu;Qiang Lu;Yunxing Wu;Zhiyuan Huang;Nanjing Hydraulic Research Institute;Dam Safety Management Center of the Ministry of water Resources;State Key Laboratory of Water Disaster Prevention;

【机构】 南京水利科学研究院； 水利部大坝安全管理中心； 水灾害防御全国重点实验室；

【摘要】 针对土石坝渗流病害演化缓慢、运行工况变化导致监测数据非平稳等问题,传统固定阈值与经验统计方法难以精准识别异常。本文结合无监督时间序列异常检测中的自适应建模思想,提出一种面向长期运行工况变化的渗流异常自适应识别模型。该模型以卷积编码结构提取监测数据特征,引入动态维度自适应表征机制刻画不同工况下的正常渗流统计关系,并通过对抗性双流结构综合重构误差与判别构建异常指数,结合指数加权移动平均与滑动分位阈值实现自适应更新与判别。工程算例表明,该方法在工况变化背景下能够保持对正常渗流性态的稳定跟踪,对异常演化过程具有较好的敏感性与鲁棒性,为土石坝渗流病害的长期监测与异常识别提供了一种具有工程可解释性的自适应分析方法。更多还原

【Abstract】 Concerning the issues such as the slow evolution of seepage-related defects in earth-rock dams and the non-stationary monitoring data caused by varying operational conditions,traditional fixed thresholds and empirical statistical methods struggle to accurately identify anomalies.To address this issue,this study proposes an adaptive seepage defect identification model oriented toward long-term operational condition changes,drawing on adaptive modeling concepts from unsupervised time-series anomaly detection.The proposed model employs a convolutional encoding structure to extract representative features from monitoring data and introduces a dynamic dimension adaptive representation mechanism to characterize the normal statistical relationship between piezometric head and reservoir water level under varying operating conditions.An adversarial dual-stream architecture is then constructed to integrate reconstruction error and discrimination scores into a unified anomaly index,which is further smoothed and adaptively updated using an exponentially weighted moving average and sliding quantile-based thresholds.An engineering case study demonstrates that the proposed method can stably track normal seepage behavior under changing operating conditions while maintaining sensitivity and robustness to abnormal evolution processes,providing an interpretable adaptive analysis approach for long-term seepage monitoring and anomaly detection in earth-rock dams.更多还原