Abstract: Safety monitoring instruments often operate in complex environments, where there exist intricate coupling relationships between measurement points, and the monitoring data may be contaminated by noise or errors to some extent. In view of the above issues in the monitoring data, an improved K-means clustering method is first applied to each sub-sequence to cluster multiple measurement points into correlated measurement point clusters. Taking into full consideration the spatial correlations of multiple measurement points, a sliding time window is set, and the Temporal Local Outlier Factor (Temporal LOF) method is used to deeply identify and remove outliers in the measurement sequences. After removing outliers from the sequences, a Bidirectional Long Short-Term Memory (BiLSTM) neural network is used to repair the measurement points with missing or abnormal sequence values. Finally, the effectiveness of the proposed method is verified using an earth-rock dam as an engineering case study.