Lithology mapping by a hybridization of the firefly and affinity propagation algorithms

The lithological identification of layers crossed by the borehole trajectory or the direct extraction of geological information from the well logs is a classic challenge in formation evaluation and has expressive impact over realistic values calculations for porosity and oil saturation. In this work we present a computer-aided interpretation of the M-N plot based on a new hybrid algorithm, which refines the affinity propagation clustering technique. For the optimization of clusters' searching, it is used a preference vector that flags affinity propagation with M-N points that are the possible best candidates to be the lithological classes representatives. These possible best candidates are obtained by a firefly metaheuristic optimization of a Kernel function, built over the M-N space. Finally, a minimum distance criterion, with respect to M-N mineral points is applied to the set of exemplary points acquired from this approach, in order to associate the lithologies. This methodology was applied in synthetic and real M-N data and in our tests, proves its effectiveness in producing a realistic lithological identification even for highly-spread M-N data.