Automated History Matching Using Combination of Adaptive Neuro Fuzzy System (ANFIS) and Differential Evolution Algorithm

Abstract

History Matching is one of the most important steps to calibrate the Reservoir Simulation models. There are hundreds of thousands of grid blocks in reservoir simulation model, therefore manual History matching is not feasible for these cases. Automatic history matching uses mathematical algorithm and techniques to adjust reservoir engineering data rather than direct engineering judgment. History matching problem is highly non-unique and in order to assess the Non-Uniqueness of the problem, Multiple Models History Matched Models are required.The History Matching with Multiple Models requires a very large number of reservoir simulations which can yield slow Field developments and large computational effort, which are very costly to make effective decision in terms of time and money. Often due to the lack of availability of multiple history matched models, there would be a lack in proper assessment of non-uniqueness, and it costs the wrong decision in terms of dry well or low productivity well. To minimize this problem some techniques such as Spline, Neural Networks, Kriging and Experimental Design, have been presented in the literature to be used as proxies to reservoir simulator. But all proxies can deliver only one history matched model, which can be highly non unique.In this paper combination of Adaptive Neuro-Fuzzy System (ANFIS) and Differential Evolution algorithm is proposed, which can minimize the above problems, by obtaining various history matched models with minimum number of simulation runs and by this way non unique problem can be minimized and accurately assessed in less time. Inputs for ANFIS model are the reservoir simulation model parameters like porosity and permeabilities of three layers and output would be differences between observed and simulated field pressure and oil production rates, water production rates of wells, which are merged together to form an objective function of error or misfit. Differential Evolution obtains Global optimum results through ANFIS, so it cuts down the expensive simulation time. The results showed the large potential of application of the technique introduced.

Submersible pumpset failure prediction using artificial intelligence methods

“…Перед началом работы необходимо подготовить исходную выборку. Для этого требуется проверить данные на наличие аномальных значений или же на несоответствия типам данных, а также при необходимости удалить строки с недостающими данными [10] Для проверки качества работы искусственного интеллекта необходимо применить метод crosstab библиотеки Pandas [3,7,15,16]. Данный метод выводит матрицу ошибок, на которой наглядно видно количество случаев верного или же неверного определения аварий [4,5,17,18].…”

Symmetric Nonlinear Feedback Control and Machine Learning for Sustainable Spherical Motor Operation

“…While AI shows promise, it may encounter limitations in certain scenarios, potentially leading to less-than-optimal results. As the field of spherical motor control advances, researchers strive to refine AI-driven strategies to overcome these obstacles, aiming for an improved control of the multi-degree-of-freedom capabilities of spherical motors [40][41][42][43][44][45][46][47][48][49][50][51].…”

Investigating the efficiency of gas re‐injection process of an oil field using combined integrated field simulation and intelligent proxy model application

“…[11] Neuro-fuzzy systems are used as a proxy model in the petroleum industry, especially in history matching. [12,13] A neuro-fuzzy system is defined as a fuzzy based system that can be trained using a training algorithm which is the main feature of neural networks. Usually they are represented as multilayer neural networks that are feed-forward.…”

“…[9,10] Therefore, the proxy models were introduced as a substitute or a help for these simulations. [12,13] A neuro-fuzzy system is defined as a fuzzy based system that can be trained using a training algorithm which is the main feature of neural networks. Instead of running numerical simulations of the reservoir, a mathematical function is defined, which is a function of input variables and returns the results of a numerical simulation.…”