Density Logging Curve Reconstruction Method Based on KNN-Transformer Algorithm

Density logging is a key technique for calculating reservoir physical parameters, identifying lithology, and evaluating oil and gas reserves. Due to factors such as borehole conditions and poor tool contact, density curves often suffer from local data gaps, distortion, or noise interference. To address these issues, this paper proposes a density logging curve reconstruction method that integrates the K-nearest neighbors algorithm and the Transformer algorithm (KNN-Transformer). The method first employs KNN to retrieve samples with temporal sedimentary characteristics similar to the target segment within a multi-dimensional logging feature space. By calculating the Euclidean distance between the target segment and historical samples across multi-dimensional features such as acoustic travel time, natural gamma ray, and resistivity, the K most similar neighboring samples are selected to construct an enhanced geological prior input set, thereby improving the geological representativeness of the input data. Subsequently, the multi-head self-attention mechanism of the Transformer algorithm is utilized to establish long-range dependencies between arbitrary positions in the depth sequence, effectively integrating local similarity constraints with global sequential patterns. This achieves a synergistic representation of local features and global structures. Experimental results show that the KNN-Transformer algorithm achieves a mean absolute error (MAE) of 0.017 0 and a coefficient of determination (R2) of 0.953 3 for density curve reconstruction. Compared to typical algorithms such as support vector regression (SVR), linear regression, and long short-term memory (LSTM), the value of MAE is reduced by 30% to 60%. The method demonstrates higher reconstruction accuracy for both the overall trend and local details of the density logging curve, along with better stability and correctness at lithological interfaces and in complex intervals. This approach effectively recovers missing sections, corrects distortions, and suppresses noise, significantly improving both the numerical accuracy and geological plausibility of the reconstructed curves. It provides a reliable technical pathway for high-quality logging data reconstruction under complex reservoir conditions.