Enhancing well log curve synthesis with selective attention long short-term memory network

In geological exploration projects, well log curves, as the primary carriers of information, are prone to data defects due to geological conditions, logging equipment, and unexpected events. This paper proposes a low-cost curve synthesis method based on deep learning. The method in this paper is based on a recurrent neural network, which can preserve contextual information in signals, crucial for logging data that vary with depth. An attention mechanism is employed to enhance the vanilla long short-term memory network, enabling it to capture larger spatial dependencies, but introducing a significant amount of matrix operations. To simplify this computation, a selector is designed to reduce the time complexity from \(O(n^{2} )\) to \(O\left( {n\log n} \right)\). Two application scenarios are considered: predicting missing logging parameters using complete logging parameters and predicting missing segments of a well based on the original well data. Through validation and analysis, the proposed method demonstrates higher accuracy. This accurate, efficient, and cost-effective prediction method holds practical value in engineering applications.