Lithology Identification in Optuna-BiGRU Logging Based on Enhanced Multi-head Attention Mechanism
Lithology identification in well logging plays a crucial role in oil and gas exploration and development.However,existing algorithmic models face challenges in effectively capturing deep-seated correlations and depth-related relationships within log curve data.They also exhibit weak fitting capabilities,struggle with accurate feature extraction,contend with noise interference,and face complexities in fine-tuning model hyperparameters.To address these issues,a model for lithology identification that combines Optuna hyperparameter optimization with a bidirectional gate cycle unit(BiGRU)and enhanced multi-head attention(EMHA)was proposed.The model is called Optuna-BiGRU-EMHA.The approach introduces residual mechanisms and layer normalization to enhance the multi-head attention module,effectively resolving issues encountered when processing well-logging data.Additionally,it leverages the Optuna hyperparameter optimization framework and wavelet packet adaptive threshold method to address hyperparameter fine-tuning and noise interference.The study begins by selecting well-logging data of five characteristic parameters,including natural gamma,deep induction resistivity,neutron-density porosity,average neutron-density porosity and lithology density,through cross-plot and sensitivity analyses.The data is then denoised using the wavelet packet adaptive threshold method and segmented into data blocks.Subsequently,the Optuna framework is employed to optimize the hyperparameters of the BiGRU-EMHA model.Finally,experimental comparisons are conducted to assess the accuracy of eight models in lithology identification based on well logging data.The eight models are KNN,RF,XGBoost,LSTM,BiGRU,BiLSTM,BiGRU-MHA,and Optuna-BiGRU-EMHA.The results show that the Optuna-BiGRU-EMHA model achieves an identification accuracy of 80％.Compared with traditional machine learning and deep learning models,there is a substantial improvement in comprehensive lithology identification accuracy by 15.94％to 23.14％,and 3.93％to 15.94％.The Optuna-BiGRU-EMHA model provides a robust theoretical foundation for conventional well-logging lithology identification in the domain of oil and gas exploration and development.
万方数据
维普
