The main aim of well log interpretation is to determine the reservoir geological and petrophysical properties downhole as well as properties fluids and their distributions in a reservoir, and identification of hydrocarbon from well logs is the most important job. Neural networks, with their great capability in adaptation and expressing arbitrary complexity, seem especially suited to solve for the reservoir problems where there widely exist uncertainty, fuzziness and nonlinearity. Much work has been published on the identification of lithology and lithofacies from well logs using neural networks. But very few, and almost no work could be found to deal with identifying hydrocarbon from well logs in this neural network approach In this paper, a supervised neural network (a multi-layer perceptron (MLP)) and two unsupervised neural networks (a self-organizing mapping (SOM) net and a fuzzy neural network (FNN) are used to identify oil from well log, and almost each of them can distinguish between oil, water, oil-water transitionand dry zones directly from well logs, but with different accuracy. A case study from a Chinese oil field is given to show their advantages and limitations, and their implementation is also given. The associated oil test results are available to demonstrate their dependability And it is shown that there is a very good agreement between results oil test and oil identification from both LP and FNN.
Generally, the main task of well log interpretation is to determine the geological and petrophysical properties of reservoir rock formation as well as properties of formation fluids and their distributions. Obviously, well logs are the fundamental information source of well log interpretation, and the eventual aim of well log interpretation in oil exploration is to determine both roperties of subsurface fluids and their distribution, which is not directly from well logs and usually based on the predetermined geological and / or petrophysical properties from well logs. During this conventional log interpretation process, it often needs selection of empirical formula and various empirical parameters, which makes the actual log interpretation extremely complicated and subjective, and eventually lead to the extreme difficulty in enhancing the chance of success of well log interpretation.
Well log interpretation, in certain viewpoint, is to invert subsurface geological and petrophysical properties, as well as properties of subsurface fluids and their spatial distributions from well logs. And well log responses measured with wireline tools, which reflect the properties of the subsurface reservoir, are in extremely complex relationship with various reservoir properties, because of the high heterogeneity and anisotropy widely existent in reservoir medium. Thus the solution of the conventional log inversion is not of singularity. So, primarily, log analysts could simply interpret well logs qualitatively, aided with their field experience obtained from success and fail by trial and error, that is, visual log interpretation of oil. Because of advances of logging technology and improvement of various experiment techniques as well as introduction of advanced scientific technology such as computers, the log analysts began to quantitatively determine both the properties of subsurface fluids and their distributions from well logs, according to various empirical formula obtained by experiments and various mathematical and physical models established on basis of theoretical analysis, which eventually made well log interpretation grow to be a largely integrated engineering science, that is, the computerized og interpretation stage started when the identi
