Artificial neural networks method for oil systems identification and its applications

We could transform partial differential equation models of some oil reservoir systems, such as well test interpretation, into series composed of nonlinear function terms. Every term was complex nonlinear function of stratigraphic parameters /spl theta/. The number of term n called as structure parameter of the model was related to model structure. We served functions as nonlinear neural units to establish function link artificial neural networks models of oil reservoir systems We applied F-test in system identification theory to determine the structure parameter n, and used multistep generalized gradient learning algorithms we developed to estimate the weighting coefficients of the networks. Stratigraphic parameters were the foundation of well test interpretation, so we requested their unique estimate values. And the problem became more difficult because aforesaid functions were multimodal functions and very sensitive in extreme point about the change of /spl theta/. Neither iteration methods nor genetic algorithms existing literatures effect. We developed a new pattern of genetic algorithms to solve above problem. It has high precision that our method is applied to build models of above systems. Moreover it can obtain unique estimate values of stratigraphic parameters.