Assisted interpretation of wireline logs

Well logs interpretation, an essential step for an efficient exploration/exploitation of the subsurface, is often a rather subjective and highly  time-consuming  process.

Our objective is to automatically extend the interpretation of well log data on a large scale with a quality comparable to what is done manually by experts on a small scale.

In the proposed method, we use a four-step process:

• a selection of the best  classification methods (supervised or unsupervised),
• an  automatic selection of the best classification result according to geological criteria,
• the introduction of an extended learning dataset for further supervised classification, and

• a combination of different supervised methods to obtain better prediction performance and a depth-dependent quantification of the quality of the prediction.

The first results on blind test data compare well with expert independent observation.

By virtue of its automatic nature, the method reduces the costs and expert intervention time and accelerates industrial processes, while guaranteeing a quality comparable to the classifications produced by the experts in the intervals identified in depth. In addition it offers a way to concentrate human work where it is most important, to manage the time and energy of experts more optimally.