Using Laplacian Support Vector Machine for Gas-bearing Reservoir Identification

Machine learning algorithms with automatic feature extraction capacity are a potent tool for breaking through the restrictions of old technologies in the new era of intelligent oil and gas exploration and have thus become widely used in geophysical exploration. When using seismic data for gas-bearing reservoir identification, a large amount of seismic data can be acquired, but it is more difficult to label the seismic data. To solve this difficulty, this research suggests using a semi-supervised machine learning method, the Laplace Support Vector Machine (LapSVM), to learn the frequency variation properties of seismic waves in gas-bearing reservoirs and subsequently identify them. The frequency variation characteristics are derived from the high and low-frequency amplitudes obtained through S-transform. The application experiments of the field data show that, in comparison to the Support Vector Machine (SVM), which only uses a small number of labeled samples for learning, the recognition based on LapSVM can use a large number of unlabeled samples for assisted learning, significantly improving recognition accuracy and enabling the effective detection of gas-bearing areas.