Reservoir characterization of F3 block (North Sea) using seismic attributes and probabilistic neural network

Hydrocarbon explorations usually are performed based on seismic inversion techniques in which there exists computational complexity. Therefore, application of simpler methods such as probabilistic neural network could be considered to decrease uncertainties of the results. The present research used a probabilistic neural network to characterize the sand reservoir of F3 block in North Sea. This algorithm applied the seismic attributes of energy, similarity and instantaneous amplitude as input parameters to estimate porosity distribution of the F3 reservoir. Calculating the likelihood probability is dependent on the smoothing parameter. Therefore, the cross validation technique was used to determine the smoothing parameter that equals to 0.21. This paper considered 16 porosity classes from 0.22 to 0.3 as output of probabilistic algorithm. This algorithm calculated the posterior probability for every point in reservoir to determine the class of each point. The maximum posterior probability was selected as the final output. The obtained results were compared with the linear equation driven regression model for acoustic impedance and porosity values. The comparison showed that the developed network could detect gas-bearing region. Also, the confusion matrix was used to validate the results and the total accuracy parameter was calculated as 0.7587 and 0.4623 for probabilistic neural network and linear regression, respectively. Therefore, Bayesian neural network could be introduced as an effective tool to explore the hydrocarbon-bearin layers because of computational complexity of seismic inversion techniques.