engleski

Lithology prediction by artificial neural networks and preparation

Well logs of spontaneous potential and of shallow and deep resistivity were pre-processed for the purpose of lithology prediction. Several artificial neural networks were trained on data from one well for the purpose of lithology prediction in a second. Data were taken from Upper Miocene intervals from two wells in the Kloštar field. Two learning approaches and three methods of prediction were applied. The results show that the best approach for inter-well lithology prediction is by training the neural network on an entire well interval, instead of training a separate neural network for each interval and subsequently employing them as an ensemble for the prediction.

artificial neural networks, lithology prediction, Miocene, Croatia

nije evidentirano