Identification and classification of water absorption profile of distributed optical fiber vibration signal based on XGBoost algorithm

Abstract

Aiming at the problem that the layered water injection cannot be accurately obtained in the process of oilfield development, this paper proposes an optical fiber vibration signal recognition and classification algorithm based on XGBoost integrated learning. Firstly, the optical fiber vibration signal collected by the optical fiber vibration sensor is denoised by variational mode decomposition and reconstruction. Then generate spectrograms corresponding to different water absorption layers at different times. A dataset containing 3000 spectral images is obtained. The data sets are imported into support vector machine, random forest classifier and XGBoost integrated classifier based on decision tree for recognition and classification. The parameters of the obtained model are optimized and the model is evaluated by cross validation. Finally, the obtained models are compared and tested. The experimental results show that the XGBoost ensemble learning algorithm with decision tree as the base classifier can effectively identify different vibration signals. This method has a certain significance for the identification of layered relative water absorption of water injection profile.