Feature extraction and classification for induced microseismic signals during hydraulic fracturing: Implication for coalbed methane reservoir stimulation

Abstract

Before effectively analyzing the stimulation of coalbed methane (CBM) reservoirs using microseismic (MS) monitoring, it is necessary to accurately distinguish signals caused by hydraulic fracturing (HF) from interference signals. In this study, the Mel-frequency cepstral coefficient-fuzzy decision tree (MFCC-FDT) signal classification method was used. To minimize the loss of crucial details during preprocessing, feature extraction is accomplished by computing the MFCC values. This is followed by a decrease in the dimensionality and fuzzification of the dataset. Finally, the preprocessed data are entered into the FDT classifier that has been trained, thereby completing the automated identification of the induced signals. The proposed technique was applied to examine MS signals during the staged HF stimulation of a CBM reservoir. These findings suggest that the MFCC-FDT method outperforms the other combinations in terms of Accuracy, Precision, Recall, and F1-Score. Thirty-five interference MS events, including signals from tunneling blasts and machine operation during reservoir stimulation, were eliminated. The total stimulated reservoir volume under the MFCC-FDT technique validation was 22 966.29 m³, 1954.34 m³ less than the volume prior to the interference signals being removed. The proposed method reveals the nonlinear frequency characteristics of the induced MS signals and can be utilized to render more accurate MS signals for quantifying CBM reservoir stimulation by subsequent source inversion.