Enhanced Early Warning Threshold Setting for Dam Safety Monitoring Based on M-Estimation and Confidence Interval Method

Accurate online identification of abnormal sudden change observations is crucial for ensuring data reliability and has been a key challenge in dam safety monitoring. Traditional methods, such as those based on the Pauta criterion, often fail to effectively identify anomalies in complex data sequences like step-type and oscillatory-type data, primarily due to unreasonable early warning threshold settings. To address this issue, this paper introduces a novel method for setting early warning thresholds by combining the scale estimator ST based on the location M-estimator with the confidence interval radius D derived from predicted values, thereby constructing the MZ criterion with a threshold of 3ST+D. The proposed model demonstrates strong resistance to outliers and good robustness, effectively improving the accuracy of online anomaly identification for various data sequences. Engineering applications have shown that the MZ criterion-based identification method achieves a low misjudgment and omission rate, high recognition accuracy, and is highly reliable for online dam safety monitoring. This method holds significant value for both theoretical research and practical engineering applications.