Innovative Approach to Assist History Matching Using Artificial Intelligence

Abstract

Abstract The study objective is to investigate the use of Artificial Intelligence (AI) methods to accelerate the history matching process. A new criterion for measuring the deviation of the simulation model from measured and /or observed parameters has been introduced. Instead of comparing parameter deviations in wells to input changes on regional basis, it is proposed to calculate a regional RMS (Root Mean Square)-error, so that the impact of input changes can be directly evaluated. Instead of grouping grid blocks based on geology, it is proposed here, to generate regions of similar trends based on all available information. Artificial intelligence (AI) is used via Self Organizing Maps (SOM) to cluster grid blocks of similar behavior. SOMs can process any kind of information; in this case these types of parameters have been particularly used:geological description: lithofacies typehydraulic flow units (HFU): permeabilities, porositiesinitialization: water saturations (initial and critical), initial pressurediscretization: spatial discretization (e.g. DZ), grid block pore volumessecondary phase movement: relative permeability endpoints A three fold approach for improving and/or assisting the history matching (AHM) work-flow using Artificial Intelligence has been tested:Use production plots, Neural Networks and "Material Balance with Interference (MBI) method for quality control and consistency check of time dependent and static data.Use the multi-dimensional cross-plot and SOM to evaluate reservoir and well performance.Use SOMs and the region RMS error to evaluate the performance of history matching runs. This new approach is simple and leads to a clear improvement of the match quality and significantly reduces the number of runs needed to achieve the match. Different field models have been used to develop this new AHM workflow. Finally in this paper, two of them are selected to demonstrate the improvement of model pressure and watercut matches using this new method. Introduction History matching a numerical simulation model is an inverse problem, which cannot be solved directly. An iterative procedure has to be applied to reduce the deviation of the model calculations and measured values. Assisted history matching allows the automation of low level processes, without taking over the key decisions from the reservoir engineer. The workflow can be divided into two categories. The first one uses gradient based optimization methods, requiring additional programming in the numerical simulation program code itself (Ref.1–7). The second category consists of algorithms and workflows, which do not require calculations inside the simulator (Ref. 8–12). The approach presented in this paper does not need the calculation of gradients. However gradients and even automated history matching tools can be used in combination with this new procedure. The creation of clusters (subsequently referred to as regions) represents a major step forward for the use of any assisted or automated history match process. The second part of the paper (weighted RMS factor) directly relates to the problem described in Ref. 17. By using the weighted RMS-error, the objective function (it quantifies the misfit between simulated and observed data) will be defined much better, discarding a lot of the otherwise possible solutions to the inverse problem. A New Approach to Assist History Matching (AHM) History matching is defined by finding a set of model parameters that minimize the difference between calculated and observed measurement values like pressure and fluid production rates. Investigating a process to leverage Artificial Intelligence to improve and speed-up history match simulation models by incorporating all reservoir and field data is the main objective of this study. A user friendly process where the simulation engineer can interact and control the parameter modifications on their own (see Figure 1) has been developed.