A novel approach for global solar irradiation forecasting on tilted plane using hybrid evolutionary neural networks

Hits: 6851
Research areas:
Year:
2021
Type of Publication:
Article
Keywords:
solar forecasting, inclined plane, artificial neural networks, evolutionary learning, hybrid algorithms, optimization
Authors:
Journal:
Journal of Cleaner Production
Volume:
287
Pages:
125577
Month:
March
ISSN:
0959-6526
BibTex:
Note:
JCR(2021): 11.072 Position: 24/279 (Q1) Category: ENVIRONMENTAL SCIENCES
Abstract:
An efficient management of solar power systems requires direct and continuous predictions of global irradiation received on inclined planes. This paper proposes a new approach that simultaneously estimates and forecasts inclined solar irradiation. The method is based on a multi-task Hybrid Evolutionary Neural Network with two output neurons: one estimates the irradiation at the current instant and another predicts it for the next hour. An Evolutionary Algorithm is used to learn the most proper topology (number of neurons and connections). Two studies are carried out to evaluate the performance of the method, considering experimental ground data for the same inclination angle and satellite data with different tilt angles. The data only contain one measured variable, what improves its applicability to other sites. The potential of three different basis functions in the hidden layer is compared (Sigmoidal Units, Radial Basis Functions and Product Units), concluding that the results achieved by Sigmoidal Units are better. Single and multi-task models are also compared with a statistical analysis, which shows no significant differences. However, the proposed multi-task option is much simpler and computationally efficient than individual models. The problem tackled is more complex and challenging than previous works, due to inclined solar irradiation is predicted based on the horizontal irradiation and also because the model simultaneously estimates and predicts irradiation. However, the performance obtained is excellent compared to the literature.
Comments:
JCR(2021): 11.072 Position: 24/279 (Q1) Category: ENVIRONMENTAL SCIENCES
Back