Perpustakaan Digital ITB

As machine learning technology continue to transform the industries with innovation that make work fast and efficient, it’s use in Oil and Gas Industries still quite limited. Applying machine learning technology in oil and gas industries enable to increase ease of operation. Lately, numerous studies present machine learning technique for identifying hydrocarbon potential zone to reduce strenuous log interpretation work, named supervised and unsupervised learning model. Supervise learning use labeled data for training, while unsupervised learning does not use any labeled data training. Most work use supervised learning model due to its simplicity and objectivity. However, this model requires lot of time and work, also often giving result in noisy data In this paper, purely unsupervised learning model is used to predict potential zone so this method does not require label for input well-log data. Well-log response from same lithology and rock type would have high similarity, so we apply an unsupervised clustering algorithm, to assign similar traits into same group and dissimilar traits into different group. Specifically, we apply k-means clustering and also we reduce dimensionality to compress data while maintain its structure and usefulness using Principal Component Analysis and t-Distributed Stochastic Neighbor Embedding algorithm. Our model is tested using LAS data from field X, one of carbonate reservoir field in Indonesia. Particularly, our clustered result shows highly similar distribution compare to commercial software clustering result, with over 80% accuracy for number of points clustered. We also validate best cluster as potential zone with further log interpretation from the commercial software. Our model thus shows good performance for zone clustering which suggest future development of any well-log clustering model and furthermore be the first step to develop hydrocarbon potential zone.