Hydrogen Leakage Location Prediction in a Fuel Cell System of Skid-Mounted Hydrogen Refueling Stations
Open Access
Abstract
Hydrogen safety is a critical issue during the construction and development of the hydrogen energy industry. Hydrogen refueling stations play a pivotal role in the hydrogen energy chain. In the event of an accidental hydrogen leak at a hydrogen refueling station, the ability to quickly predict the leakage location is crucial for taking immediate and effective measures to prevent disastrous consequences. Therefore, the development of precise and efficient technologies to predict leakage locations is vital for the safe and stable operation of hydrogen refueling stations. This paper studied the localization technology of high-risk leakage locations in the fuel cell system of a skid-mounted hydrogen refueling station. The hydrogen leakage and diffusion processes in the fuel cell system were predicted using CFD simulations, and the hydrogen concentration data at various monitoring points were obtained. Then, a multilayer feedforward neural network was developed to predict leakage locations using simulated concentration data as training samples. After multiple adjustments to the network structure and hyperparameters, a final model with two hidden layers was selected. Each hidden layer consisted of 10 neurons. The hyperparameters included a learning rate of 0.0001, a batch size of 32, and 10-fold cross-validation. The Softmax classifier and Adam optimizer were used, with a training set for 1500 epochs. The results show that the algorithm can predict leakage locations not included in the training set. The accuracy achieved by the model was 95%. This approach addresses the limitations of sensor detection in accurately locating leaks and mitigates the risks associated with manual inspections. This paper provides a feasible method for locating hydrogen leakage in hydrogen energy application scenarios.Keywords
Funding Information
- Technology Project of State Grid Corporation of China (5419-202219075A-1-1-ZN)
This publication has 24 references indexed in Scilit:
- Hydrogen leakage location prediction at hydrogen refueling stations based on deep learningEnergy, 2023
- Real-time plume tracking using transfer learning approachComputers & Chemical Engineering, 2023
- Optimized model‐based diagnosis approach for hydrogen leakage in hydrogen supply system of fuel cell truckInternational Journal of Energy Research, 2022
- Leak localization using distributed sensors and machine learning for hydrogen releases from a fuel cell vehicle in a parking garageInternational Journal of Hydrogen Energy, 2021
- Leak detection systems in oil and gas fields: Present trends and future prospectsFlow Measurement and Instrumentation, 2020
- Global economic and environmental outcomes of the Paris AgreementEnergy Economics, 2020
- Hydrogen production, storage, transportation and key challenges with applications: A reviewEnergy Conversion and Management, 2018
- Numerical study of hydrogen release accidents in a residential garageInternational Journal of Hydrogen Energy, 2015
- Dispersion and burning behavior of hydrogen released in a full-scale residential garage in the presence and absence of conventional automobilesInternational Journal of Hydrogen Energy, 2012
- A logical calculus of the ideas immanent in nervous activityBulletin of Mathematical Biology, 1943