A New Approach Combining Neural Networks and Simulated Annealing for Solving Petroleum Inverse Problems

A wide range of petroleum problems can be solved by using global optimization algorithms such as simulated annealing method (SAM). This paper combines state of the art neurocomputing with SAM to solve a complex inverse problem in a novel way. The automatic estimation of relative permeability and capillary pressure curves from laboratory corefloods is used to illustrate the use of a neural network, the development of a new training backpropagation algorithm, the use of fuzzy logic for system identification, and finally the use of simulated annealing in an automatic history matching algorithm where the numerical simulator is replaced by two fast neural networks.