Using Probabilistic Unsupervised Neural Method for Lithofacies Identification
Salim Chikhi and Mohamed Batouche
Computer Science Department, University Mentouri of Constantine, Algeria
Abstract: This paper presents a probabilistic unsupervised neural method in order to construct the lithofacies of the wells HM2 and HM3 situated in the south of Algeria (Sahara). Our objective is to facilitate the experts' work in geological domain and to allow them to obtain the structure and the nature of lands around the drilling quickly. For this, we propose the use of the Self-Organized Map (SOM) of Kohonen. We introduce a set of labeled log’s data in some points of the hole. Once the obtained map is the best deployed one (the neuronal network is well adapted to the data of the wells), a probabilistic formalism is introduced to enhance the classification process. Our system provides a lithofacies of the concerned hole in an aspect easy to read by a geology expert who identifies the potential for oil production at a given source and so forms the basis for estimating the financial returns and economic benefits. The obtained results show that the approach is robust and effective.
Keywords: Lithofacies, differed well logging, self-organized map, probabilistic formalism, classification, underground cores.
Received October 4, 2003; accepted January 3, 2004