SLOPE DEFORMATION DATA CALIBRATION BASED ON LINEAR REGRESSION AND SEGMENTED IDENTIFICATION ALGORITHMS

This study develops an algorithmic framework for slope deformation monitoring data calibration and staged deformation identification. The selected modeling section focuses on correcting systematic deviation in fiber-optic displacement measurements and identifying transition nodes in surface displacement sequences. A univariate linear regression model is established between fiber-optic displacement data and vibrating-wire displacement reference data, and the calibration parameters are solved by minimizing the residual sum of squares. Five-fold cross-validation is used to evaluate stability, while MSE, RMSE, MAE, and R2 measure calibration accuracy. The full-sample fitting result reaches MSE = 7.195, RMSE = 2.682, MAE = 1.308, and R2 = 0.999, indicating that the calibrated displacement closely follows the reference sequence. For staged deformation identification, the method combines Savitzky-Golay smoothing, Hampel outlier detection, sliding Welch t-test screening, physical duration and monotonicity constraints, and segmented fitting. The final transition nodes are identified at indices 5261 and 8861, dividing the sequence into slow uniform deformation, accelerated deformation, and rapid deformation. The average velocities of the three stages are 0.1753, 0.6115, and 4.1211 mm/h, and the fitted R2 values are 0.947975, 0.983713, and 0.998814. The results show that the proposed framework can reduce sensor bias, identify deformation-stage evolution, and support monitoring-based warning analysis.