Research Based On Machine Learning Method For Seismic Event Detection

The events picking is an important part of seismic data process,and the accuracy of events picking is directly related to the processing result of speed analysis and normal moveout and static correction in next steps.In the process of seismic field data processing,whether it is the ultra-deep data in the western of China or the complex geological structure data,the target events signal,especially the microseismic signal,is often hidden in the background field noise interference,so it is difficult to identify the signal with traditional methods.This paper uses one method of machine learning to solve the problem of events detection.After searching the boosting algorithm group,I choose the gradient boosting decision tree algorithm as the basic training model.I use numerical simulated shot record which the seismic model forward,and design a complete feature extraction including mathematical statistic,signal analysis and image processing which can character the seismic attributes parameter.After getting the feature from seismic record,I use the analysis of variance and the support vector machine for the feature selection,then we put the selected feature in the GBDT model for training.This paper uses the thin interbed model forward shot record with random noise,55,200 pieces of training data used to train,verify and test.The precision of the events picking in the test set reaches 92%,which verified the effectiveness of the GBDT method.Moreover,I test the data forwarding by the Marmousi model with random noise,and the precision is 94%.I also analyze the single trace extracted from the record and compare it with different signal to noise ratio.The GBDT algorithm used in this paper demonstrates its excellent performance in the identification of the events picking,such as algorithm accuracy,calculation efficiency,and universality of parameter adjustment.It provides a new idea for the events picking,and reflects the great potential of machine learning algorithms in the field of seismic data process.