Reservoir characterization and porosity classification using probabilistic neural network (PNN) based on single and multi-smoothing parameters

Abstract

A probabilistic neural network (PNN) is a feed-forward neural network using a smoothing parameter. We used the PNN algorithm based on single and multi-smoothing parameters for multi-dimensional data classification. Using multi-smoothing parameters, we implemented an improved probabilistic neural network (PNN) to estimate the porosity distribution of a gas reservoir in the North Sea. Comparing the results of implementing smoothing parameters obtained from model-based optimization and particle swarm optimization (PSO) indicated the efficiency of PNN in characterizing the gas. Also, results showed that while the PSO algorithm was able to specify smoothing parameters with more precision, about 9%, it was very time-consuming. Finally, multi PNN based on PSO was applied to estimate the porosity distribution of the F3 reservoir. The results validated the main fracture or gas chimney of the F3 reservoir with higher porosity. Also, gas-bearing layers were highlighted by energy and similarity attributes.

Keywords

• probabilistic neural network
• smoothing parameter
• model-based optimization
• particle swarm optimization