Enhancing 1D Liquid Pipeline Temperature Modeling With Online Learning and Optimization

Abstract An innovative approach has been developed in the form of transient thermal modeling, which is finely tuned as an online learning and optimization problem. Extensive testing and validation have demonstrated the effectiveness of this approach in fine-tuning temperature profile across a wide spectrum of pipelines and operations. This method integrates Dakota software from Sandia lab with the Synergi Pipeline Simulator (SPS) from DNV to carry out a derivative-free global optimization. Real-time learning is applied to an external pipe heat source and soil conductivity, using a loss function defined as the difference between the modeled and measured temperatures. The mean loss is continuously employed to adjust the values of soil conductivity and the pipe heat source. The theoretical framework can be seamlessly implemented in other pipeline simulator packages, providing a versatile solution for the industry.