Perpustakaan Digital ITB

The oil and gas industry was already under immense pressure following the 2014 oil price crash. The outbreak of COVID-19 exacerbates the already egregious conditions by dropping demand in an unprecedented manner. Nevertheless, due to its role as one of the biggest affordable energy suppliers, its existence will still be paramount for decades to come. With decreasing profit margins in the E&P business, optimizing existing fields may be the best strategy moving forward. One of the alternatives is identifying and developing low resistivity/low contrast (LRLC) reservoirs which can significantly increase reserves and production with minimum outlay. However, since these zones are hard to distinguish through conventional log analysis, they are often bypassed. Luckily, it has been shown that machine learning algorithms can improve efficiency and help solve many problems in the petroleum industry. A novel systematic workflow, augmented with machine learning, was developed to rapidly identify and interpret potential LRLC zones to tackle this problem. LRLC zones can be defined as commercial hydrocarbon producing intervals without the useful contrast in measured resistivity with shale and/or water-bearing zones. The occurrence of LRLC zones may be caused by laminated shale-sand sequences, conductive minerals, fine-grained sands, and microporosity. In all those cases, significant amounts of hydrocarbons have been reported to be produced. First, a thorough analysis of the major causes of LRLC zones and their respective manifestations in the logging curves was conducted. Unsupervised learning algorithms were then used to cluster the log data based on their similarities. Consequently, potential LRLC zones can be identified efficiently. Combined with all available data (e.g., mud logs, core analyses, etc.), the cause(s) of the LRLC pays could be identified, and water saturations could be computed more accurately. The developed workflow was then applied to 15 wells in the Gumai formation for validation. Comparisons with the results obtained from conventional log analysis, done by a third party, showed that the developed program could identify all LRLC zones, including potential zones still bypassed by the conventional analysis. This workflow will help log analysts to rapidly identify and interpret LRLC zones, which could significantly increase efficiency, reduce risks, drive profits, and create value in stringent times.