Velotopic maps in well-log inversion

The use of an unsupervised Kohonen neural network is proposed for estimating the basic parameters of the formation surrounding a well. After proper learning, the neural network is transferred into two maps directly labeled in terms of the shear and compressional velocities of the formation. They are denoted as velotopic maps. The estimation method based on them consists of the following steps: (1) process by a batch Burg predictor of order 20 the acoustic log: (2) apply the vector constituted by the reflection coefficients, after normalization, to the neural network; (3) determine the most excited neuron; and (4) consider, as estimates, the values of the velocities associated to the previous neuron in the velotopic maps