Permeability prediction from well logs using an evolutionary neural network

Reservoir permeability estimation from well logs is a difficult task for petrophysicists. Use of neural networks, in recent years, has shown great promises over many traditional techniques. However, the conventional learning algorithms (e.g. backpropagation) do not give satisfactory results when the solution is trapped in local minima. This paper presents an application of an evolution algorithm that allows trapped networks to achieve the optimal solution. The proposed technique is applied to two examples from China and Indonesia. The results show that the evolutionary network has a better chance in finding the optimal solution in much shorter time. The log-derived permeability estimations are realistic and geologically meaningful.