Method of predicting sandstone fracture process based on improved VMD and GA-BP neural network

In order to extract effective acoustic emission signal characteristics of sandstone fracture and improve the accuracy of predicting sandstone fracture process, a prediction method based on improved variational mode decomposition (VMD) algorithm and GA-BP neural network was proposed. Firstly, uniaxial compression experiments were carried out to simulate the process of sandstone fracture, and the AE signals of fracture process were collected; secondly, in order to obtain effective AE signals, the effective characteristic parameters were extracted for prediction, the correlation coefficient was introduced to improve VMD algorithm, the original AE signals were preprocessed, and the signal energy characteristic parameters were extracted as the input of the model so as to distinguish fracture process. Finally, the GA-BP prediction model was constructed, the weights and thresholds of BP neural network were optimized by genetic algorithm (GA), and the signal energy was used as a sample to train the prediction model. The results showed that the introduction to correlation coefficient could effectively solve the problem that K value was difficult to select in VMD algorithm, and the collected acoustic emission signals could be effectively de-noising; in addition, the BP neural network prediction model improved by GA algorithm could accurately predict the fracture state; compared with the traditional BP neural network model before the improvement, the stability of the improved one was higher, with better convergence ability, and prediction accuracy increased by 17.5%。