Well-log correlation using a back-propagation neural network

We present a back-propagation neural network with an input layer in the form of a tapped delay line wich can be trained effectively on one or several well logs to recognize a particular geological marker. Subsequently, the neural network proposes locations of this marker on other wells in the field. Another neural network, similar in architecture to the first one, performs the same task for secondary markers using, in addition to the well logs, a depth reference function to the first marker. This method is shown to have better performance and better discrimination than standard crosscorrelation techniques. It lends itself well for an interactive implementation on a workstation.