Artificial neural networks for short-term forecasting of solar-driven green hydrogen production

The main objective of this study was to investigate the performance of a solar PV powered electrolyser for green hydrogen production for renewable electricity generation to meet the electrical load of buildings in desert regions and the development of short-term artificial neural networks (ANN) forecasting models to predict ahead green hydrogen production. The goal was to develop a clean and sustainable power system with 100% renewable resources and alternative zero carbon fuels for future net zero energy buildings. The proposed hybrid solar PV/fuel cell system was used to meet the electrical loads of an educational building under harsh desert weather conditions (hot and humid conditions - effects of dust, temperature, and humidity). Modelling, simulation analysis, and optimization were carried out in this study to investigate in detail the technical, economic, and environmental performance of the system. The input and output data from the simulation analysis were used as input in the Artificial Neural Networks prediction model to forecast ahead the hydrogen production from the electrolyser using solar energy. The results showed that the proposed stand-alone hybrid solar PV/fuel cell power system met the daily (1599 kWh/day), monthly, and yearly electrical load of the building without power shortage. The yearly electrical production from the hybrid solar PV /Fuel cell power system showed that 79.7% (1,239,774 kWh/year) and 20.3% (316,346 kWh/year) were generated respectively from the solar PV and fuel cell. The total yearly energy input in the electrolyser was 898,415 kWh/year and the hydrogen production in the electrolyser was 19,360 kg/year while the fuel cell fuel consumption was 15,817 kg/year. The daily average and maximum green hydrogen production in the electrolyser were 2.21 and 10.8 kg/hr respectively with a specific energy consumption of 46.4 kWh/kg. For the ANN-based forecasting models, the R values obtained for the training, validation, and test for the solar PV power output and green hydrogen production were respectively 0.9665 – 0.97424 and 0.97983 – 0.98187. The values of ANN regression model correlation R and the mean square errors MSE showed that the ANN models created were very accurate in projecting the power output from solar PV system and green hydrogen production from the electrolyser.