Well logging super-resolution based on fractal interpolation enhanced by BiLSTM-AMPSO

Enhancing the level of geological characterization and analysis has always been a challenging task in the exploration and development of unconventional oil and gas reservoirs. In order to address these challenges, geophysical logging is one of the most important data for characterizing target reservoir model. However, the vertical resolution of conventional well logging data is often insufficient to handle the fine-scale characterization tasks of complex unconventional reservoirs. Therefore, several vertical resolution enhancement techniques have been devised to attain as fine as possible reservoir characterization. Nevertheless, local sharp structure with sufficient high frequency information cannot always been well recovered. In this paper, to improve the vertical resolution of well-logging data, a novel fractal interpolation based well logging super-resolution method was proposed by employing bidirectional long short-term memory (BiLSTM) and adaptive mutation particle swarm optimization (AMPSO). Specifically, mutation factors are introduced into the particle swarm optimization (PSO) algorithm to enhance search accuracy. The AMPSO is utilized to obtain optimal solutions for the vertical scaling factors in the fractal interpolation iterated function system (IFS), ensuring these factors adhere to fractal theory constraints. This approach constructs an IFS that aligns closely with the fractal characteristics of the logging curves, thereby enabling more effective extraction of structural information and finer-scale information recovery. Experimental results indicate that, compared to well logging super-resolution methods such as bicubic interpolation, convolutional neural networks, and random forests, the proposed method not only effectively preserves the overall contours and detailed information of the well logging curves but also exhibits good stability across different regions and various well logging curves.