dGB Releases OpendTect 6.6
Along with its Machine Learning Plugin , the first fully integrated E&P Machine Learning platform
The platform combines OpendTect's extensive seismic interpretation system with the R&D world of Python, TensorFlow, Keras, and Scikit Learn. Trained Machine Learning models can be deployed directly on both seismic and well data. This allows operational geoscientists to benefit directly from all the latest Machine Learning algorithms without needing programming skills. The environment also allows you to develop new workflows and train models to add value to your data. This is done with the existing tools and models available in the OpendTect platform. Newly trained models can be shared with others ensuring rapid deployment in operational settings. Researchers can accelerate their Machine Learning research by programming in a Python environment that takes care of data IO and visualization. Deep learning models created in this environment are directly available in the Machine Learning Plugin for further use.
For seismic predictions, the first Machine Learning release supports Convolutional Neural Networks and Autoencoder-Decoders implemented in Keras (TensorFlow). A trained U-Net (Autoencoder-Decoder) fault predictor model is included in the software. This model transforms 3D seismic data to fault probability volumes. Fig. 1 shows the result of the U-Net fault predictor after application of a thinning algorithm. For log-to-log and seismic-to-log predictions the Machine Learning platform offers a range of cutting-edge algorithms from Scikit Learn, including: linear regression, multi-layer perceptrons, support vector machines, and ensemble methods including Random Forests, Adaboost and Gradient boosting. Fig. 2 shows an example of predicting Pore Volume from DT, RhoB, RD and GR logs with eXtreme Gradient boosting of Random Forest Models.