No abstract

No abstract

Kharyaginskoe field is located on the North part of Russia and produced by Zarubezhneft-Production Kharyaga company. The field is on the mature stage, so, it requires the special attention to drilling and production strategies optimization. The field is highly complicated by geological nonconformities (for example, faults), which has the crucial impact on the well productivity, recovery and reservoir management decision. So, the comprehensive understanding the zones complexities such as connectivity and compartmentalization is vital for optimal field development and has always been extremely challenging. The traditional method of using formation pressures for evaluation of reservoir connectivity does not perfectly work for the reservoirs under production as the fluid extraction disturbs the initial pressure equilibrium. This study shows another possible methodology of reservoir connectivity examination based on the analysis of hydrocarbon composition for three zones of Kharyaginskoe field.

The Kharyaginskoye oil field is located on the territory of the Nenets Autonomous District and belongs to the Timan-Pechora Basin oil and gas province. The main object of development is a Devonian age carbonate reservoir. The productive zones of the studied object are mainly confined to thin bed low-porosity reservoirs with a complex structure of void space. The high heterogeneity of deposits laterally and the presence of different levels of oil-water contact (OWC) in the marginal isolated zones necessitate a more accurate assessment of the oil-saturated effective thicknesses. The increase in the reliability of the interpretation was achieved by the joint analysis of borehole and seismic studies using Machine Learning methods. At the stage of configuring the facies model based on well logs and core data, a Multi-Resolution Graph-based Clustering MRGC was used, which provides effective integration of geological and geophysical information. The multi-dimensional dot-pattern recognition method based on k-Nearest neighbors algorithm (k-NN), and by combining various criteria, it allows solving the problem of non-linearity of the relationships between logging responses and the corresponding lithology. The algorithm of the democratic association of neural networks DNNA was used to propagate electrofacies in the inter-well space. The method optimizes the use of seismic data before summation and after summation together with well data through a controlled process that provides a calibrated and scaled distribution of facies. The most probable facies distribution can be used directly as a property in reservoir modeling or as a constraint for modeling. It is known that there is no direct connection between a certain type of wave pattern and the lithological composition of rocks, therefore, the analysis of changing reflection characteristics is performed in conjunction with geophysical data, such as well logging. In addition, a priori geological information about the work area is involved. An important condition for the effective application of facies analysis is the presence of representative core material and the availability of high-quality well information. At the first stage of the work, the lithotyping of carbonate deposits was performed according to the macro description of the core, based on the classification of limestones according to R. H. Dunham. Then, using the multidimensional statistical recognition algorithm MRGC, the relationships between the selected lithotypes and logging responses were obtained. As a result of the tuning, a cluster model was obtained that allows us to distinguish electrofacies characterized by an increased filtration and capacitance potential. At the second stage, the obtained electrofacies, considering the nature of saturation, were used to train cubes of seismic attributes and calculate the cubes of lithofacies and the probability of the existence of each lithofacies. The key point in the distribution was the use of electrofacies obtained in wells belonging to different facies zones. Thus, the joint analysis of all available borehole and seismic information by machine learning methods made it possible to make a forecast lithofacies considering the type of saturation based on geological and geophysical information analysis. The effectiveness of the presented technologies was demonstrated by analyzing the properties of low-permeable carbonate reservoirs, where classical attributes and inversion demonstrate limitations in describing a heterogeneous saturation model. The use of neural network approaches allows to configure complex nonlinear dependencies that are not available to classical methods. The use of a small volume of multi-scale geological and geophysical information using Machine Learning algorithms in the field of field-geophysical and seismic interpretation makes it possible to increase the reliability of interpretation and clarify the location of prospective zones with improved reservoir properties on the studied area, as well as to minimize geological risks during subsequent well placement.