The Use of Neural Network Technologies in Prediction the Reservoir Properties of Unconsolidated Reservoir Rocks of Shallow Bitumen Deposits

In the conditions of the dynamically changing conjuncture of the oil and gas market, there is an urgent need to reduce the cost of oil production and increase the efficiency of development, this is especially important for the local ultra-viscous oil. In this regard, it is necessary to optimize costs at all stages, starting from the geological exploration and even at the stage of completion of the development process. For ultra-viscous oil deposits, this is especially relevant at the stage of assessing the resource potential of a separate uplift of any of the fields, when the only reliable way to perform a high-frequency section at shallow depths is to drill appraisal wells with full core sampling. An additional load is exerted by the period between core extraction and obtaining information about the flow properties of each of the samples. By themselves, standard core studies are complicated by the fact that sand rocks of weakly cemented bitumoids can often be destroyed during experiments. In this regard, the use of new approaches, including digital ones, which allow us to make quick decisions on a part of the geological section in the area of the appraisal well and on the uplift as a whole, are highly in demand. This article describes the methods that allow the determining of flow properties for uncemented (loose sands) rocks in Permian sediments. More than 25,000 core samples were studied from 805 wells at several fields of the Republic of Tatarstan. The technology used allows us to calculate a continuous curve of volumetric bitumen saturation in the conditions of complete or partial absence of core at the well. This paper presents the results of creating an algorithm for automatic prediction of weight bitumen saturation in a sand pack of the Sheshminsky horizon of the Permian system using neural network technologies, as well as using an alternative calculation method.