Looking for a fracture plausibility using wireline log data.
It is adapted after Aguilerra (1995) on "Naturally Fractured Reservoir" book. A method is developed to detect fracture using wireline log by calculating its possibility in depth. Logs are categorized into five categories:
- Electrical category
- Multi-pad category
- Radioactivity category
- Rugosity category
- Acoustic category
Plausibility are calculated for each one category using "criterion" which then combined as a final plausibility using Bayesian logic. In this code, I already attached some calculation to carry out this work. However, the result of Bayesian logic to calculate final plausibility seems not reasonable in this case because it leads to a very high probability on almost all depth. An element in the logic includes several fraction multiplication which yield a very small number, it might not represent the true plausibility. Thus I tried to introduce a cross entropy in this code to calculate the final plausibility. Moreover, I equipped the algorithm with neural net using PyTorch in case you already have fracture data and want to perform a supervised learning through it.
Notes: Pb: probability using Bayesian logic Pce: probability using cross entropy Proba: probability using neural net
I also made it in Google Colab :D If you're interested feel free to check
Next: I want to improve using fracture porosity
Any kind of suggestions are welcome!