This article presents the methodology involving the combined use of machine learning elements and a physically meaningful filtration model. The authors propose using a network of radial basis functions for solving the problem of restoring hydraulic conductivity in the interwell space for an oil field. The advantage of the proposed approach in comparison with classical interpolation methods as applied to the problems of reconstructing the filtration-capacitive properties of the interwell space is shown. The paper considers an algorithm for the interaction of machine learning methods, a filtration model, a mechanism for separating input data, a form of a general objective function, which includes physical and expert constraints. The research was carried out on the example of a symmetrical element of an oil field. The proposed procedure for finding a solution includes solving a direct and an adjoint problem.

No abstract

The method of cyclic waterflooding is one of the most cost-effective methods of enhanced oil recovery. Simulation of cyclic waterflooding will take much more time than conventional waterflooding. This is due to the time step limitation which should be much less than the half-period of fluctuations and the characteristictime of pressures equalization in neighboring interlayers. This paper proposes the flow equations averaged over the cycle period in the case of a periodic law of variation in production rates or bottomhole pressures in wells that are free from the specified time step limitation and allow to simulate cyclic waterflooding in a time comparable to the time of simulating a conventional waterflooding. The simulations based on the averaged and conventional (used in flow simulators) equations of a two-phase flow are compared using a synthetic example and actual field areas in Western Siberia and Kazakhstan. A complete solution to the cyclic waterflooding problem involves the selection of the promising areas, well locations, as well as the period and duration of stimulation. As a rule, this requires running multiple simulations. Therefore, in order to reduce the number of simulation runs, in this paper we propose an efficient technology for 3D modeling of cyclic waterflooding in oil fields. This technology allows to select, in real time, the promising areas and to assess the effect on incremental oil recovery of such factors as locations of wells, the frequency of cycles, etc. Also, it allows to predict incremental oil production in a time much shorter than using a conventional simulator, without accuracy losses. The successful results of technology application in some fields of Western Siberia and Kazakhstan are given below.

No abstract