Integrating artificial intelligence based hybrid deep learning prediction models to estimate exergy efficiency for realistic solar photovoltaic power plants: validation with ground measurements

Ms Almas, Sivasankari Sundaram,
Integrating artificial intelligence based hybrid deep learning prediction models to estimate exergy efficiency for realistic solar photovoltaic power plants: validation with ground measurements,
Thermal Science and Engineering Progress,
Volume 65,
2025,
103902,
ISSN 2451-9049,
https://doi.org/10.1016/j.tsep.2025.103902.
(https://www.sciencedirect.com/science/article/pii/S2451904925006936)
Abstract: Though energy efficiency quantifies the instantaneous power generation, it does not embody the true operational output of a photovoltaic (PV) system, which brings exergy efficiency into limelight. Current model approaches available for estimation of exergy efficiency of PV system baselines thermodynamics-based approach. This involves model assumptions, a greater number of model constants and are dependent on attributes which cannot be easily accessed or priorly estimable. So, this paper develops a robust dual layer and multi-layer hybrid based deep learning, artificial intelligence models that predict exergy efficiency with priorly estimable inputs. This minimizes complex modelling and optimization challenges in solar PV system. A training data set for a long-term two-year annual duration of 191.9 kWp grid-dependent solar PV plant located in Khopoli, India is considered. The prediction is made considering four different easily estimable and proposed inputs like irradiance, wind speed, module temperature and ambient coefficient (A). The performance of all the hybrid models was found appropriate with Mean Absolute Percentage Error (MAPE) ranging from a minimum of 1.88 % to a maximum of 1.99 %, during training. Further, all the estimators are validated for a monitored operation of PV plants at Telangana and Bengaluru. Among the proposed model’s dual layer based GRU + BiLSTM and multi-layer hybrid model compete for these validated locations with a least MAPE of 1.81 % and 1.88 % for Telangana & Bengaluru respectively. So, the usability and behaviour of the proposed models against the traditional physical model stands appropriate, acceptable and provides a way forward for performance optimization.
Keywords: Exergy efficiency prediction; Multi-layer hybrid model; Dual-layer hybrid model; Realistic PV plant; Deep learning