Application of Genetic Algorithm in Deconvolution of Logs

Abstract

Abstract The wave pattern observed during data acquisition from the subsurface rock is often affected by heterogeneity, formation fluid, tool resolution and other forms of noise. Often however, the aim of the engineer or geologist is to generate the original pattern by deconvolving the wave such as in seismic interpretation. Subsequently, several models have been designed over the years but most of them require the use of cumbrous logical sequences which are difficult to understand and are thus written in computer programs. This paper aims to solve the problem by the use of a method that involves the backward modelling of such wave patterns using genetic algorithm. This works on the principle of evolution of data such that different individuals converge to a perfect individual, imitating natural evolution. Using this method, a model was created to deconvolve waves. The assumption made was that some waves (including logs) originate from two extreme divergent points before the influence of ‘noise’. The new model was used to recreate the wave pattern and was compared to the input wave. The model was tested on pseudo-logs wave patterns and the results were estimate to be between 80% and 95% accurate. The significance however, is that such model can be used to recreate lithological patterns which can be used to project forward models by applying certain rock properties.