A Method For Predicting Well Logs Using Bi-LSTM Based On Correlation Analysis

Well log can reflect the formation lithology and stratigraphic characteristics,and how to predict and repair well logs is of great value in application.However,current well logging prediction methods use the existing log data directly without considering the correlation between the data and use only unidirectional networks,that is,without the use of a Bi-directional Long Short-Term Memory(Bi-LSTM).In this thesis,we propose a new method for logging curve prediction based on Bi-LSTM,which takes into account the correlation between logging curves and the correlation between different logging curves,and uses a bidirectional LSTM model instead of a unidirectional LSTM model to predict well log,which significantly improves the accuracy of logging curve prediction.In this thesis,we propose the use of Bi-LSTM model to predict logging curves,which can learn and generate prediction models faster and more comprehensively by taking into account the back-and-forth correlation between logging curves using bi-directional prediction model,and also further improve the prediction accuracy.The results of this study show that the prediction effect of Bi-LSTM model is significantly better than that of LSTM model.It proposes to combine the logging curve correlation analysis with the Bi-LSTM model,and select the data with higher correlation as the input of the model through the logging curve correlation analysis,which reduces the use of sample data to a certain extent,and also greatly improves the prediction accuracy.Its research results show that the prediction accuracy will improve with the increase of correlation between the logging data,which has a good application prospect in the field of oil logging.