Phase equilibrium analysis in recovery and transportation of natural hydrogen

Abstract

In order to optimize the recovery and long-distance transportation technology of the emerging energy resource, natural hydrogen, the complex thermodynamic properties of the natural hydrogen production mixture is focused in this paper, aiming at improving the development, purification and pipeline transportation operation of natural hydrogen based on the multi-component phase equilibrium analysis. Commencing with the second law of thermodynamics and following the criteria of of reducing the total energy and increasing the total entropy in a system, a thermodynamic-consistent flash calculation method is formulated by using the Onsager reciprocal relations to construct a time-dependent evolution scheme of mole composition and volume over time. The dissipative nature of the algorithm's energy behavior is substantiated. Based on this iterative flash calculation method, a thermodynamics-informed neural network is developed. Environmental temperature is extracted as the environmental parameter of the input features based on thermodynamic analysis, along with the consideration the total molar concentration. The critical temperature, critical pressure, acentric factor, and molar fraction of each component are designated as the thermodynamic characteristics for each component to be employed as the remaining input features. The output layer of this network will acquire the outcomes of phase stability testing, specifically, the total number of phases in the fluid at equilibrium and the results of phase partition calculation, which encompasses the compositional mole fraction in each phase at equilibrium. These predictions are highly valuable for the rapid determination of phase equilibrium states in a multi-component system. An iterative flash evaporation algorithm is employed to acquire an ample dataset for training machine learning models and tuning parameters of neural network architectures. A self-adaptive deep learning algorithm and corresponding deep neural network configurations are formulated, with the integration of advanced machine learning techniques, to enhance the training and validation performance. By improving the training efficiency and prediction accuracy of the deep learning algorithm, this accurate and fast phase equilibrium prediction method can be used to analyze the distribution of phase equilibrium states under different compositions of natural hydrogen products. Additionally, it offers practical process recommendations for hydrogen purification and the separation and storage of carbon dioxide. The fast phase equilibrium prediction model is coupled with the intelligent regulation and control system of gas transmission pipelines to achieve intelligent detection of pipeline operation safety and intelligent regulation to meet the demand for gas transmission supply in natural gas pipeline containing admixtures.

Keywords

• new energy
• thermodynamics in multi-phase flow
• hydrogen energy
• multi-component multi-phase flow
• machine learning
• pipeline dispatching