基于半监督学习的烃源岩成熟度预测

1、Prediction of Source Rock Maturity Using Semi Supervised (EAGE2020)

unlabeled data is easier to obtain but unsupervised learning algorithms are limited to uncovering the underlying pattern in the data and feature reduction, and results need then to be interpreted a posteriori.

Method

The workflow suggested in Figure 1 illustrates the commonly known data processing stages required for a machine-learning project.

Conclusions

The model can be improved by the addition of labels and highly correlating attributes, such as inversion results.

In the case of high uncertainty in the labels, we recommend using a combination of models to understand the uncertainty and reduce operational risk.