Parameters corresponding to different algorithms.

In engineering structure performance monitoring, capturing real-time on-site data and conducting precise analysis are critical for assessing structural condition and safety. However, equipment instability and complex on-site environments often lead to data anomalies and gaps, hindering accurate performance evaluation. This study, conducted within a wind farm reinforcement project in Shandong Province, addresses these challenges by focusing on anomaly detection and data imputation for weld nail strain, anchor cable axial force, and concrete strain. We propose an innovative iterative rolling difference-Z-score method for anomaly detection and a machine learning-based imputation framework combining linear interpolation with LightGBM. Experimental results show that the iterative rolling difference-Z-score method detects single-point and clustered anomalies with a Z-score threshold of 4, achieving robust performance even with 80% data loss. The imputation framework maintains low mean squared error (MSE) of 0.0214–0.0227 and root mean squared error (RMSE) of 0.14–0.15 for continuous missing data scenarios (60–200 points), with reliable reconstruction up to 50% data loss. This research provides a robust solution for ensuring the precision and integrity of wind farm monitoring data, enhancing long-term structural reliability in renewable energy applications.