Denoising methods are a highly desired component of signal processing, and they can separate the signal of interest from noise to improve the subsequent signal analyses. In this paper, an advanced denoising method based on a fully convolutional encoder–decoder neural network is proposed. The method simultaneously learns the sparse features in the time–frequency domain, and the mask-related mapping function for signal separation. The results show that the proposed method has an impressive performance on denoising microseismic signals containing various types and intensities of noise. Furthermore, the method works well even when a similar frequency band is shared between the microseismic signals and the noises. The proposed method, compared to the existing methods, significantly improves the signal–noise ratio thanks to minor changes of the microseismic signal (less distortion in the waveform). Additionally, the proposed methods preserve the shape and amplitude characteristics so that it allows better recovery of the real waveform. This method is exceedingly useful for the automatic processing of the microseismic signal. Further, it has excellent potential to be extended to the study of exploration seismology and earthquakes.

SUMMARY: Denoising and onset time picking of signals are essential before extracting source information from collected seismic/microseismic data. We proposed an advanced deep dual-tasking network (DDTN) that integrates these two procedures sequentially to achieve the optimal performance. Two homo-structured encoder–decoder networks with specially designed structures and parameters are connected in series for handling the denoising and detection of microseismic signals. Based on the similarity of data types, the output of the denoising network will be imported into the detection network to obtain labels for the signal duration. The procedures of denoising and duration detection can be completed in an integrated way, where the denoised signals can improve the accuracy of onset time picking. Results show that the method has a good performance for the denoising of microseismic signals that contain various types and intensities of noise. Compared with existing methods, DDTN removes the noise with a minor waveform distortion. It is ideal for recovering the microseismic signal while maintaining a good capacity for onset time picking when the signal-to-noise ratio is low. Based on that, the second network can detect a more accurate duration of microseismic signals and thus obtain more accurate onset time. The method has great potential to be extended to the study of exploration seismology and earthquakes.

The degree of rock mass discontinuity is crucial for evaluating surrounding rock quality, yet its accurate and rapid measurement at construction sites remains challenging. This study utilizes fractal dimension to characterize the geometric characteristics of rock mass discontinuity and develops a data-driven surrounding rock classification (SRC) model integrating machine learning algorithms. Initially, the box-counting method was introduced to calculate the fractal dimension of discontinuity from the excavation face image. Subsequently, crucial parameters affecting surrounding rock quality were analyzed and selected, including rock strength, the fractal dimension of discontinuity, the discontinuity condition, the in-situ stress condition, the groundwater condition, and excavation orientation. This study compiled a database containing 246 railway and highway tunnel cases based on these parameters. Then, four SRC models were constructed, integrating Bayesian optimization (BO) with support vector machine (SVM), random forest (RF), adaptive boosting (AdaBoost), and gradient boosting decision tree (GBDT) algorithms. Evaluation indicators, including 5-fold cross-validation, precision, recall, F1-score, micro-F1-score, macro-F1-score, accuracy, and the receiver operating characteristic curve, demonstrated the GBDT-BO model’s superior robustness in learning and generalization compared to other models. Furthermore, four additional excavation face cases validated the intelligent SRC approach’s practicality. Finally, the synthetic minority over-sampling technique was employed to balance the training set. Subsequent retraining and evaluation confirmed that the imbalanced dataset does not adversely affect SRC model performance. The proposed GBDT-BO model shows promise for predicting surrounding rock quality and guiding dynamic tunnel excavation and support.

Due to the complexity of the various waveforms of microseismic data, there are high requirements on the automatic multi-classification of such data; an accurate classification is conducive for further signal processing and stability analysis of surrounding rock masses. In this study, a microseismic multi-classification (MMC) model is proposed based on the short time Fourier transform (STFT) technology and convolutional neural network (CNN). The real and imaginary parts of the coefficients of microseismic data are inputted to the proposed model to generate three classes of targets. Compared with existing methods, the MMC has an optimal performance in multi-classification of microseismic data in terms of Precision, Recall, and F1-score, even when the waveform of a microseismic signal is similar to that of some special noise. Moreover, semisynthetic data constructed by clean microseismic data and noise are used to prove the low sensitivity of the MMC to noise. Microseismic data recorded under different geological conditions are also tested to prove the generality of the model, and a microseismic signal with M_w ≥ 0.2 can be detected with a high accuracy. The proposed method has great potential to be extended to the study of exploration seismology and earthquakes.

The high-stress hazards of underground engineering have stimulated the exploration of microseismic monitoring and early warning methods. To achieve a good monitoring effect, the monitoring object is usually enclosed by a microseismic array (sensor array) (e.g., slope engineering, etc.). However, some characteristics of a buried tunnel, including “linear”, “deep-buried”, and “long”, make it difficult to deploy a reasonable microseismic array, which leads to the microseismic array being non-enclosed for the monitoring object. Application of the non-enclosed microseismic array yields decreases the accuracy of the source location. To solve the problem wisely, this paper deals with the feasibility of non-enclosed microseismic arrays (axial-extended, lateral-extended, and twin-tube arrays) by introducing a quantitative method. To this end, an optimized microseismic array with the best source location accuracy for a twin-tube expressway tunnel is proposed. The obtained results reveal that the non-enclosed microseismic arrays, which are unavoidable in expressway tunnel engineering, do not introduce errors but reduce the ability to resist them. Further, the twin-tube array achieves a better source location accuracy than the axial and lateral-extended arrays. In the application of the source location based on the particle swarm optimization (PSO) algorithm to the twin-tube array, microseismic events, which cluster in the rockburst section, are wholly gathered, and the maximum error is reduced by about 30–50 m, indicating its greater feasibility with respect to the single-tube array.