Seismic intelligent characterization of paleocave filling types in deeply buried carbonate reservoirs

Abstract

As carbonate rocks are susceptible to diagenesis, it can form large paoleokarst systems which enrich more than one billion tons of hydrocarbon at depth exceed 6000 m in Tarim basin. However, due to the influence of multi-stage sedimentary filling and coalesced collapsion, the spatial distribution of effective reservoirs is extremely complex. Affected by the low-pass filtering effect by the deep strata, the main frequency of seismic data is about 25 Hz. It’s difficult for conventional technologies, such as seismic attributes, spectral decomposition, and seismic waveform clustering to effectively predict the development location of unfilled caves, which seriously restricts the fine exploration and efficient exploration of paleokarst reservoirs. The core-logging-seismic nonlinear regression model has been established by machine learning algorithm, guiding high precision identification of electrofacies labeled by core data and reservoirs characterization of seismic facies labeled by well logging feature parameters, forming a set of 3D space multi-layered reservoir characterization technology that synthesizes multi-scale geophysical data: 1) Reservoir recognition of electrofacies: Using the manifold learning algorithm of Uniform Manifold Approximation and Projection (UMAP), two characteristic parameters which reflect the reservoir lithology and physical features were obtained by nonlinear dimensionality reduction of conventional logging data. The geological labels of sample points were colibrated by core data, which include surrounding rock, paleocave sedimentary, paleocave collapsed breccias, fracutres etc. Furthermore, their relative threshold values in low-dimensional space were extracted. 2) Nonlinear mapping of electrofacies to seismic facies: Spectral shaping and diffusion filtering algorithms were used to obtain spread spectrum and denoised seismic data, enhancing the domain frequency of the origional seismic dataset. Since different filling modes can cause transverse waveform disturbances, the variance tendencis of waveforms were used to calculate the maximum likelihood extrapolation of feature parameters based on Bayesian framework. Two 3D seismic datasets with lithology and physical features were generated. 3) Spatial geometric structure characterization of paleocave fillings: The results of single-point classification in the two-dimensional crossplot of characteristic parameters were mathematically counted, and the thresholds representing the surrounding rock, karst cave filler, and filling type were delineated. Accordingly, 3D high-precision depiction of reservoir distribution and filling structure had been calculated by 3D characteristic parameter datasets. 4) 3D geological interpretation: Based on the high- precision identification results, the spatial distributions of the paleokarst reservoirs were interpreted by the sequence stratigraphy and paleokarst geology theory. The filling structure and composition relationship of the cave interior was described, deriving 3D modeling of multi-scale deeply buried fractured-vuggy carbonate reservoirs. The method was applied to the research area of 76 km2. The conventional logging data of 23 wells had been used to recognize collapsed breccias, sedimentary fillings, etc. Based on high resolution seismic data, a wide range of interlayer paleokarst reservoir and paleocave filling types were characterized. This application can break through the bottleneck of reservoir characterization determined by a single factor of seismic impedance. It organically integrated geological theory with geophysical data, providing a new technic for multi-scale data fusion to achieve high-precision geological characterization. Note: This paper was accepted into the Technical Program but was not presented at IMAGE 2022 in Houston, Texas.