A Collaborative Framework Based on an Improved Adaptive Cubature Kalman Filter for Multi-Anomaly Mitigation in Bridge Temperature Monitoring Data

Long-term bridge temperature monitoring data are often affected by random noise, outliers, and sensor drift, which may reduce the reliability of structural thermal-response analysis. This study proposes a collaborative framework based on an improved adaptive cubature Kalman filter (IACKF) for multi-anomaly mitigation. First, the process- and observation-noise covariance matrices are updated online using innovation and residual statistics to suppress time-varying noise. Second, a dual-Gaussian contaminated observation model is incorporated to develop an outlier-resistant improved adaptive cubature Kalman filter (OR-IACKF) for isolated and patch-type outliers. Third, an improved particle swarm optimization–backpropagation–IACKF (IPSO-BP-IACKF) scheme is used to predict a drift-free reference from adjacent monitoring points and recursively estimate sensor drift. The framework is evaluated using simulated data, constant-temperature chamber measurements, field-monitored bridge data, and a jointly contaminated dataset. The results indicate that the staged framework improves the mitigation of different anomaly types while maintaining relatively stable performance under different parameter settings. The proposed method provides a practical data-processing approach for long-term bridge structural health monitoring.