Tomographic Velocity Images by Artificial Neural Networks

Djarfour Noureddine¹, Baddari Kamel¹, Djeddi Mabrouk²

(1) Université de Boumérdes, Boumérdes, Algeria (2) Université de Boumérdes, Boumérdes, Algeria

In order to obtain velocity image from a borehole to borehole seismic tomographic experiment, the artificial neurone networks of Elmen type, were trained to reconstruct the velocity from the traveltime. This type of network offers an advantage of training simplicity by the Back-propagation conjugate gradient algorithm. The behavior observed of networks on training data is very similar to the one observed on test data. The efficiency of these networks is tested with the complex geologic model, and the results were very encouraging. A comparison with algorithms ART and SIRT was made, and the superiority of networks of neurons was noted.