Artificial intelligent application for predictive maintenance in low emissions hydrogen production systems: Promotion of Scientific, Technological and Innovation Culture in Smart Cities
-
-
[1] Universidad de Huelva
Universidad de Huelva
Huelva, España
-
- Localización: Actas de las VII Jornadas ScienCity 2024: Fomento de la Cultura Científica, Tecnológica y de Innovación en Ciudades Inteligentes / José Manuel Lozano Domínguez (ed. lit.), Estefanía Cortés Ancos (ed. lit.), Manuel Joaquín Redondo González (ed. lit.), Tomás de J. (ed. lit.), Mateo Sanguino (ed. lit.), Iñaki Josep Fernández de Viana González (ed. lit.), Miguel Ángel Rodríguez Román (ed. lit.), 2024, ISBN 9798266036024, págs. 31-34
- Idioma: inglés
- Texto completo no disponible (Saber más ...)
-
Resumen
Hydrogen has been identified as a solution for the long-term energy storage, and it can be produced by electrolysis and renewable energy supply. To ensure electrolysers work effectively there is a need to monitor their operation in an intelligent manner to determine in advance, potential abnormalities. This paper presents a hybrid artificial intelligence (AI) concept comprising a deep reinforcement learning (DRL) and long short-term memory network (LSTM) for predictive maintenance in electrolysers plants. Once it is identified the critical variable that provide information about the plant’s state of health, the DRL algorithm searches for the set of variables within the electrolyser process data to select the one with the highest correlation to this critical variable (hydrogen flow temperature). The DRL identified that the cooling water temperature has the highest correlation coefficient (0.99) with hydrogen flow temperature. This data is then fed into the LSTM to predict hydrogen temperature with a root-mean-squared error of 0.1351. The predicted hydrogen temperature sensor value can be used to control or shut down the electrolyser plant in the event of failure of the actual sensor.
