Multi-scale seismic impedance inversion based on Transformer model and deep learning

Abstract In this paper, we delve into the field of improving the accuracy and efficiency of seismic data processing through innovative applications of multiscale seismic impedance inversion techniques, underpinned by the transformative power of Transformer model architectures and the complexity of deep learning approaches. By crafting a deep learning model that seamlessly integrates the Transformer model framework with multi-faceted feature extraction strategies, we have achieved a breakthrough in accurate inversion of seismic wave impedance in complex geological landscapes, marking an important step forward in this field. In the experiments, we used a large dataset containing more than 1000 seismic data samples covering sandstone layers of varying thicknesses (3–15 m). After meticulous training and rigorous testing, our proposed method has surpassed conventional seismic impedance inversion methods, achieving a notable surge in accuracy. Precisely, the average error margin between the retrieved impedance data and the authentic logging data has been trimmed from 15% down to 12%. This improvement is particularly pronounced in the prediction of thin sand bodies with thicknesses ranging from 5 to 10 meters, where accuracy enhancements of up to 25% have been observed. This outcome not only validates the potency of Transformer models and deep learning technologies in tackling intricate seismic data processing challenges but also underscores the pivotal role of multi-scale analysis in elevating inversion precision to new heights. We also evaluated the computational efficiency of the model. Through optimization algorithm and hardware acceleration technology, the method in this paper achieves faster inversion speed while maintaining high accuracy, and the average single inversion time is shortened to 60% of that of the traditional method. This improvement is of great significance to improve the real-time and efficiency of seismic exploration. The multi-scale seismic impedance inversion technology based on Transformer model and deep learning proposed in this paper performs well in data analysis and practical applications, bringing new solutions and technical means to the field of seismic exploration.