I2AM: a Semi-Automatic System for Data Interpretation in Petroleum Geology

The natural complexities of petroleum reservoir systems continue to provide a challenge to geoscientists. In petroleum geology, exploration and production wells are often analysed using image logs and the use of all the available borehole data to completely characterize the reservoir potentials and performance is an important task. The development of reliable interpretation methods is of prime importance regarding the reservoir understanding and data integration is a crucial step in order to create useful description models and to reduce the amount of time necessary for each study. Artificial intelligence, data mining techniques and statistical methods are widely used in reservoir modelling, for instance in prediction of sedimentary facies. The aim of our work was to define and implement a suite of tools for interpretation of image logs and large datasets of subsurface data coming from geological exploration. This led to the development of IAM (Intelligent Image Analysis and Mapping), a semi-automatic system that exploits image processing algorithms and artificial intelligence techniques to analyse and classify borehole data. More in detail, the objectives of the IAM approach are: (1) to automatically extract rock properties information from all the different types of data recorded/measured in the wells, and visual features from image logs in particular; (2) to identify clusters along the wells that have similar characteristics; (3) to predict class distribution over new wells in the same area. The main benefits of this approach are the ability to manage and use a large amount of subsurface data simultaneously. Moreover, the automatic identification of similar portions of wells by hierarchical clustering saves a lot of time for the geologist (since he analyses only the previously identified clusters). The interpretation time reduces from days to hours and subjectivity errors are avoided. Moreover, chosen clusters are the input for supervised learning methods which learn a classification that can be applied to new wells.