Abstract:

Fine reservoir description has always been the focus of development and production of unconventional oil and gas resources, but the vertical resolution of conventional logging curves is difficult to satisfy the effective identification of thin layers at centimeter or even millimeter scale. Aiming at this problem, this paper proposes a two-level knowledge migration super-resolution method for large-scale difference in logging curves to improve the vertical resolution of logging curves, thus realizing high-resolution target reservoir fine description in low-cost cases, using integrated machine learning as the basic tool and the perspective of multi-view and multi-scale as the core. The microsphere resistivity, natural gamma ray, and acoustic time difference curves with better formation response are selected as the target curves, and the construction of a mapping model from the information of high-resolution imaging resistivity curves to the target logging curves is realized, which in turn realizes the large-scale difference super-resolution of target logging curves , and the super-resolution results are compared with different super-resolution methods. The results show that the correlation coefficients between the super-resolution curves obtained by the method of this paper and the real high-resolution curves are greater than 0.9, which are improved by 3.6% to 16.0% compared with the comparison methods, and the mean square error,the root mean square error, the mean absolute error, the mean absolute percentage error, the symmetric mean absolute percentage error  are reduced by 28.9% to 90.8%, 15.7% to 69.8%, 24.4% to 74.7%,  25.0% to 74.2%, and 25.2% to 77.4% , respectively. Therefore, the method of this paper is able to largely realize the millimeter-level super-resolution processing of the existing conventional logging curves, and the obtained super-resolution curves are able to roughly capture the formation changes, which alleviates the difficulty of the problem of effective identification of fine reservoirs.

Key words: fine reservoir description, logging curves, integrated machine learning, large-scale, super-resolution