Prediction of pressure evolution with artificial neural network using extended datasets for non-venting liquid hydrogen tanks

Abstract Nowadays, storage of liquid hydrogen (LH 2 ) has become a promising solution to many energy applications compared to conventional fuels. However, self-pressurization phenomenon due to heat leakage into LH 2 tanks still represents a bottleneck for its development. Accordingly, accurate prediction of the pressure evolution is of great importance and urgent requirement for safe storage. Storage of LH 2 in a tank is a complex multi-physics problem with accompanied thermodynamic phenomena and complicated fluid-fluid and solid-fluid interactions resulting from heat and mass transfers. Due to such complications, thermodynamic homogeneous model (THM) cannot alone capture the complex hidden features and hence cannot provide accurate predictions of the pressure evolution. Herein, artificial neural network (ANN), as an intelligent experimental-based approach will be hybridized with THM to give a whole picture for LH 2 self-pressurization phenomenon using extended datasets from THM. It was found that the ANN-THM hybrid model has achieved maximum improvements of over 20 % and 90% in the average prediction accuracy compared to the individual normal ANN and THM models, respectively when applied to some literature LH 2 tanks. The integration of artificial intelligence techniques based on experimentations into the theoretical basis from thermodynamic models holds great potential as a promising avenue for future research.