순차적 인공신경망 기법을 적용한 시추위치 최적화 연구

Abstract

ABSTRACT

A Study on Well Placement Optimization Using a Sequential Method of Artificial Neural Networks

Kim Yu Mi Advisor : Prof. Jang Il Sik, Ph.D. Department of Energy & Resources Engineering Graduate School of Chosun University

Well placement is an important task to minimize the risk of unproductive drilling and to maximize the production. When dealing with heterogeneities, the intuitive engineering judgment may not be sufficient, and the use of global optimization methods becomes necessary in finding a favorable production plan. The behavior of scenario combining thousands of production locations are predicted through reservoir simulations. Depending on the size of the reservoir, the computation time and cost of the reservoir simulation may be excessive. Soft computing techniques can reduce the time and cost of solving problems. Among the soft computing techniques, artificial neural network is capable of high-speed computation while ensuring reliability of prediction results. In conventional drilling location optimization studies, artificial neural network is frequently used due to fast operation speed, but the prediction result is sensitive to the learning data and input/output data used for the learning. While it is possible to identify values close to the global optimum using the optimization algorithm, there is a limit that it is not easy to obtain the global optimum itself. In this paper, a new method is proposed by applying sequential neural networks to obtain true global optimization for well placement. By repeatedly generating artificial neural networks for the high ranked data among the results from a sequence of artificial neural networks, the range of the search space is gradually narrowed down. As a result, the global optimum point can be found through execution of minimum number of reservoir simulation. In order to evaluate the reliability of the developed method, the new method was applied for two different reservoir cases. It was confirmed that the proposed method of well location optimization can reduce the computation time by at least 70% according to the number of grids and the number of data used in learning. The proposed method was compared with particle swarm optimization (PSO) algorithm, which proved that the proposed method was more efficient than PSO. If the drilling location optimization of the on-site reservoir is performed by the method proposed in this paper, it is possible to save the time and cost by high-speed operation and obtain the global optimum infill drilling location with the maximum accumulated production.