Research On History Matching Under Facies Constraints Based On Dimensional Reduction Techniques

Automatic history matching utilize optimization theory, automatically adjuest the formation physical properties according to the difference between the historical performance and simulation data by means of computer technologies. Autmatic history matching aims at revealing true geological situation of the reservoirs. But the parameters to be adjusted are limited to continuous parameters such as permeability, porosity, saturation; gradient methods based on adjoint model which is commonly used both internationally and domestically are difficult to solve and not suitable for site application; both of the matching accuracy and the speed needs to be improved. The paper established a minimized mathematical model for reservoir simulation history matching based on Bayesian statistical theory, and then utilize the multi-realizations from geological stochastic modelling and the Singular Value Decompostion method to reduce the parameters to be adjusted and enhance computation speed. Correponding to the solution, this paper striks a balance between the implementation difficulty and computation speed, and selected and improved the parallel Simutanneous Stochastic Pertubation Optimization Algorithm, a gradient-free method.This paper extends the range of parameters: relative permeability curve and acquifer are included, what is more, sedimentary facies distributions as discrete parameters are also included. Because discrete data does not confom to Bayesian Gaussian assumption, this paper introduces the level set theory and try to inverse facies distrbution while keeping its geological continutity. In order to reduce the multi-solutions of inverse problems and to improve the matching accuracy, we propose a two-step automatic history matching process: firstly realize facies inversion, and then update the continuous geological parameters based on the updated facies model and its constraints. For large-scale reservoirs, the use of coarse meshing in facies inversion can save a lot of time. Based on the above research, this paper developed a set of automatic history matching software, including the model pre-treatment, the main matching module and results viewer and output backup module.The results showed that: the proposed history matching method based on Bayesian theory and data dimensional reduction applying gradient-free SPSA method, can effectively grasp the reservoir prior information and gives out a reasonable model parameter estimation; the proposed facies inversion method can give a reasonable facies distribution while ensuring geological continuity; the two-step automatic history matching process, not only match the accumulated historical performance with high accuracy but also give out good matches to individual well dynamic. When applied to large-scale reservoirs, the matching efficiency can be greatly improved.