A back propagation artificial neural network application in lithofacies identification

The traditional lithofacies identification by the geology method usually has the flaw of strong subjectivity, high randomness and high requirement of the geological interpreter. In order to accurately identify the lithofacies in each well and thus provide a better guidance for the plan of further exploration and exploitation, an integrated lithofacies identification method based on ANN (artificial neural network) is presented. Take the YJ Oilfield as an example, on the basis of well log data preprocessing, choose appropriate training samples and identify the lithofacies in single well taking advantage of the generalization and self-learning ability of the ANN algorithm, and compare the result to the lithofacies identification from core data. It shows that this method has relatively high accuracy when applied to lithofacies identification of clastic reservoirs which normally have complicated lithological sequences. Compared to the traditional identification method, the ANN method avoid the subjectivity in the well log interpretation and don´t have to set up interpretation models for the district which usually calls for abundant experience; what´s more, the interpreter could make a balance between accuracy and efficiency by shifting the neuron number of the hidden layer. In general, this method is of high practical value.