Hybrid Dynamic Surrogate Modelling for a Once-Through Steam Generator

Abstract

Four surrogate modelling techniques are compared in the context of modelling once-through steam generators (OTSGs) for offshore combined cycle gas turbines (GTCCs): Linear and polynomial regression, Gaussian process regression and neural networks for regression. Both fully data-driven models and hybrid models based on residual modelling are explored. We find that speed-ups on the order of 10k are achievable while keeping root mean squared error at less than 1%. Our work demonstrates the feasibility of developing OTSG surrogate models suitable for real-time operational optimization in a digital twin context. This may accelerate the adoption of GTCCs in offshore industry and potentially contribute towards a 25% reduction in emissions from oil & gas platforms. © 2023 Elsevier B.V. Author keywords Digital Twin; Gaussian Process Regression; Neural Networks; Residual Modelling; Surrogate Modelling

Hybrid Dynamic Surrogate Modelling for a Once-Through Steam Generator