Research on deep-learning-based data completion methods for fiber optic monitoring using SVM-PCA

Distributed fiber optic monitoring technology has gained widespread application in geotechnical engineering monitoring due to its advantages of high precision, long-distance capability, and continuous testing.However, complex geological conditions and construction disturbances frequently cause damage to fiber optic monitoring lines, resulting in missing monitoring data and thereby compromising the integrity and reliability of monitoring results.To effectively address this issue, this paper proposes a deep learning-based data completion method for fiber optic monitoring data that combines Support Vector Machines with Principal Component Analysis (SVM-PCA).Using measured data from the floor of a mining face in a specific mining area as the research subject, this study analyzed the characteristics of data gaps and constructed a corresponding completion model. The model was then compared and analyzed against methods such as linear interpolation, spline interpolation, and Kriging interpolation.The results indicate that for fixed continuous missing data, the SVM-PCA method achieved an average root mean square error (RMSE) of 16.07, representing reductions of 33.87%, 50.97%, and 22.91% compared to linear interpolation, spline interpolation, and Kriging interpolation, respectively. The average deviation was 13.46, representing reductions of 27.17%, 54.09%, and 20.33% compared to linear interpolation, spline interpolation, and Kriging interpolation, respectively. For randomly continuous missing data, the SVM-PCA method achieved an average root mean square error of 16.24, representing reductions of 10.49%, 49.22%, and 19.07% compared to linear interpolation, spline interpolation, and Kriging interpolation, respectively. The average deviation was 14.28, showing reductions of 45.94%, 17.57%, and 9.88% compared to linear interpolation, spline interpolation, and Kriging interpolation, respectively. Overall, it outperformed traditional interpolation algorithms.