Well logging curve reconstruction based on Genetic Neural Networks

Abstract

A technique to reconstruct wireline logs based on the Genetic Neural Networks (GNN) optimization is presented in this paper. Using the genetic algorithm to optimize the traditional neural network's topology structure, weight and threshold, it can effectively overcome traditional neural network's shortcomings of redundancy structure, tendency to fall into local minimum, etc. Firstly, taking the reconstruction of acoustic logging curve as an example, the optimized structure and relevant parameters of GNN was determined, and the validity of the technique was demonstrated. Then, in order to verify its feasibility, using the field data of two wells from land and offshore respectively, more curves including acoustic, resistivity and density logs were reconstructed. It shows that the reconstruction results in the two wells by GNN are superior to those by traditional neural network, and the agreement between the synthesized data and the expected output is better for the well 1 than that for the well 2 in this case. From the error function of curve reconstruction, it is also shown that the computing error of the genetic neural network is less than that of the traditional neural networks, which indicates that the genetic neural network is not only feasible but also has positive practicality and versatility in logging curve reconstruction.