Dept of Electrical Engineering, University of Cape Town, South Africa
10.22109/jemt.2023.413268.1465
Abstract
For optimal electricity distribution and effective management of the solar power system, the time-series hourly forecasting of the 1global solar radiation (GSR) and 2substation electric power output (SEP) based on a machine-learning (ML) are real-time information. In this solar study, two ML architectures (network models) based on the multivariate input-features for the forecasting of the hourly GSR and SEP outputs are proposed and evaluated. Our research methodology describes the historical moving data sequence processes with the ML algorithms for the non-linear modeling and multistep forecasting of the: 1GSR and 2SEP outputs. Furthermore, the ML forecast skills are evaluated with an independent dataset of 5–energy buildings. The studied findings reveal that the proposed ML architectures are sufficient (deployable) for solving a non-linear regression problem within the solar power forecast framework at any potential site. For different forecast horizons, the mean-error scores of 0.495–2.102 kW (RMSE) and 0.236–1.238 kW (MAE) are reported by the two ML models. The Conv3D model with the high-variance input features reported a high forecast skill for the most of the considered time horizons and outperformed the Conv2D network model in the: - 1solar modeling and 2multistep ahead forecasts of 0–2.5 days horizon. For large-scale integrations of the solar photovoltaic (PV) systems into the power grid-networks, our experimental findings revealed that the ML architectures with the moving sequence datasets are implementable in the real-time scenarios.
Olaofe, Z. (2023). Machine-learning models for hourly forecasts of the global solar radiation and substation electric power outputs. Journal of Energy Management and Technology, (), -. doi: 10.22109/jemt.2023.413268.1465
MLA
Zaccheus Olaofe. "Machine-learning models for hourly forecasts of the global solar radiation and substation electric power outputs". Journal of Energy Management and Technology, , , 2023, -. doi: 10.22109/jemt.2023.413268.1465
HARVARD
Olaofe, Z. (2023). 'Machine-learning models for hourly forecasts of the global solar radiation and substation electric power outputs', Journal of Energy Management and Technology, (), pp. -. doi: 10.22109/jemt.2023.413268.1465
VANCOUVER
Olaofe, Z. Machine-learning models for hourly forecasts of the global solar radiation and substation electric power outputs. Journal of Energy Management and Technology, 2023; (): -. doi: 10.22109/jemt.2023.413268.1465