Logging curve prediction method based on CNN-LSTM-attention

Logging curves are an important basis for geological development planning and hydrocarbon reserve exploration. However, in the actual logging process, there are often problems such as instrument malfunction, improper human operation and signal interference, resulting in missing or distorted logging data in a certain section. In this work, the Pearson correlation coefficients are calculated based on the actual logging data, providing empirical evidence for the correlation between the logging curves through statistical methods. Based on this, we propose a missing logging curve prediction method combining attention mechanism (Attention), convolutional neural network (CNN) and long short-term memory neural network (LSTM), and design logging curve prediction experiments and log interpretation calculation experiments. The results show that compared with the prediction results of the conventional LSTM, the absolute mean error (MAE) of the CNN-LSTM-Attention model is reduced by 57.86%, the root mean square error (RMSE) is reduced by 56.27%, and the correlation is increased by 5.40%. The constructed model has excellent performance in the prediction of logging curves. In addition, the predicted porosity of the formation interpreted by the CNN-LSTM-Attention model has less error than the true porosity calculated from the original data, and the predicted curve contains the geological characteristics of the original curve, indicating that the prediction method can be used in the field of logging interpretation.