Publications

Increasing gas demand has been significant in Indonesia and its surrounding regions in the last decade, affecting the exploration of gas resources. However, the conventional gas reserves are decreasing and becoming more difficult to find. Hence, alternate gas sources are expected to use to fulfill gas demand in the near future. Coalbed methane (CBM) is an unconventional gas whose main composition is methane. Methane is stored in the coal matrix and formed during the coalification process. CBM production needs a dewatering process as a start-up. In CBM reservoirs, water is produced from the cleats inside coal layers. The water production decreases reservoir pressure and helps coalbed methane release from coal matrix when the reservoir pressure reaches to a gas desorption pressure. The released methane flows through the micropores to the natural fractures in coal and then to wellbores. The period of water producing before the methane release is defined as a dewatering time. Predicting dewatering time is very important to CBM projects because the length of the period would affect projects’ value and it is a key to efficient gas production from CBM reservoirs. To predict dewatering time, reservoir engineers use reservoir simulation occasionally. Halim (2013) developed an equation to estimate the dewatering time as a function of reservoir area, reservoir thickness, fracture porosity, fracture permeability, desorption pressure, reservoir pressure, and bottomhole pressure. This study is to develop a system to predict dewatering time on a vertical CBM well by using Artificial Neural Network (ANN) technique, a simple procedure that would eliminate tedious and less accurate estimation. The ANN technique is a powerful modeling tool to establish the complex relationship between input parameters and dewatering time. In this study, 335 sets of data points were used for generating, validating, and testing the ANN model. The ANN system for dewatering time prediction provided an correlation coefficient R2 of 0.99 and average data error of 6.01% compared with the targeted values.