Research on High-precision Density Prediction Method Based on SSA-XGBoost

[Background]: Complex lithology well sections demand high precision in density logging data, which traditional computational models fail to adequately meet. [Purpose]: This study aims to improve the precision of density logging curves utilizing machine learning regression prediction models. [Methods]: Firstly, Monte Carlo N-Particle transport code of dual-detector density logging tool instrument was utilized to obtain stratigraphic data of varying density, which served to validate the predictive effectiveness of the model. Then, SSA algorithm was adopted to enhance XGBoost, resulting in the development of the SSA-XGBoost density prediction model. By optimizing the parameters of SVR, RFR, and LSTM employing the SSA, the SSA-SVR, SSA-RFR, and SSA-LSTM models were constructed to predict the simulated formation density. The predictive performance of each model was compared and analyzed applying quantitative evaluation metrics and Taylor diagram models. Finally, the performance of different prediction models was evaluated on actual density logging data. [Results]: In the comparative analysis and processing of actual well density logging data with various models, the SSA-XGBoost model exhibited smaller errors between predicted and actual density, demonstrating high density accuracy and validating the precision of the method. [Conclusions]: The SSA-XGBoost model demonstrates higher predictive accuracy than traditional spine-ribs plot and other models, showing great potential for applications in the processing of actual density logging data.