Machine Learning & Optimization

Machine Learning

The Machine Learning Utility (MLU) is new in Petro-SIM 7.7. It makes hybrid modelling straightforward and accessible to all Petro-SIM users, with minimum need for data science skills.

• MLU provides a user-friendly interface and workflow to generate synthetic data from KBC first principles models, configure and train ML models.
• ML Hub (Python environment) in which the ML models are trained – this is shipped with and launches seamlessly from Petro-SIM. No experience of Python programming language is needed.
• ML model configuration interfaces, where the user can set feature and target variables, hidden layers and number of nodes, activation functions, variable types and rules to generate physics-informed neural network (PINN) models
• ML training algorithms – included in ML Hub and accessed at the touch of a button.
• Base Delta unit op – extended to ingest ML models created by the MLU so they can be used within the Petro-SIM simulation platform.

For more details, please see Petro-SIM Help ML Utility section

Optimization

Petro-SIM 7.7 offers an additional 4 optimization algorithms:

• BOBYQA
• COBYQA
• COBYLA PRIMA
• SLSQP

In addition to IPOPT, COBYLA and NOMAD. The choice of optimizer algorithm will depend on the type of optimization problem which is being solved. More details can be found in the Petro-SIM help under Optimiser Engines.