“…The extracted vibration data on the X and Y axesfor feature extraction. In this study, a deep network architecture is used for the seismic data regression prediction, as shown in Fig.5, referred to references(6,7) , the network architecture is composed of one Bi-LSTM layer(8,9) (output dimension is 250, Dropout is 0.2), one LSTM(10) layer (output dimension is 100, Dropout is 0.2), and three fully connected layers (two hidden layers, neuron number 50 Dropout is 0.2, and number of neurons is 100 Dropout is 0.2), and output layer (number of neurons is 2(X and Y axes)*1000). This network model is a regression model and the scenario for its input and output data are described as follows.Set the parameter time_step=s for Bi-LSTM, and assume that the current vibration feature vector is π at time π , and the historical vibration feature vectors π , π , β¦ , π , π are the input vector sequence of the Bi-LSTM model.…”