Proceedings of the 5
th
NA International Conference on Industrial Engineering and Operations Management
Detroit, Michigan, USA, August 10 - 14, 2020
© IEOM Society International
Kernel Density Estimation of Solar Radiation and Wind Speed for
South Africa
Thand’uxolo Kenneth Magenuka, Kabeya Musasa, Kayode Timothy Akindeji
Faculty of Engineering and the Built Environment
Department of Electrical Power Engineering
Durban University of Technology
Durban, South Africa
magenuka@mut.ac.za; musasak@dut.ac.za; timothya@dut.ac.za
Abstract
Accurately determining the amount of daily solar irradiance is of paramount significance before commencing on any solar energy
projects. Similarly, precise approximation of wind speed probability distribution is essential and significant important in renewable
applications and as a result a study was performed. The paper offers a nonparametric density estimation technique for solar
irradiance and wind speed probability distribution. In literature, several probability distribution functions (pdfs) are tested both for
solar irradiance and wind speeds and compared with the proposed nonparametric kernel density estimation method. To judge the
performance and correctness of the appropriate modelling distributions, the root mean square error (rmse) and mean bias error
(mbe) are used as performance test criteria for pdfs. The results firstly demonstrate that the proposed nonparametric kernel density
estimator gives more accurate estimation with better adaptability than the commonly used conventional parametric distribution for
both solar and wind. Moreover, the study shows that the commonly used Gaussian and Epanechnikov kernel methods were the
most adaptable methods for all stations. This study will play an in important role in the country as the first-hand information in
prediction of future renewable projects.
Keywords:
Statistical distribution, kernel density estimator, root mean square error, mean bias error.
1. Introduction
With Africa’s potential to renewable energy investment and fast growing renewable energy technology, Africa has a significant share
in energy production in the future. This will assist in minimizing the high dependence on traditional ways of energy generation and
in turn help with job creation. Currently, the composition of energy mix in South Africa shows a great dependence on fossil fuels
with approximately 80% of the country’s electricity produced from coal. South Africa is relatively in an infant stage in terms of
renewable energy systems however, it is growing and according to the country’s medium-long term goals set in 2010, renewables
will account about 17 500 MW of the total energy mix by 2030. About 7 000 MW of this renewable target is expected to be phased
in and operating by the end of 2020. Moreover, solar energy seems to have a potential for massive roll out for both small scale and
large scale in many regions of South Africa and thereby having a significant share in energy production in the future.
Renewable energy system has a major role to play in alleviating this high reliance on fossil fuels and other disadvantages as discussed
according to department of energy. In modelling these renewable energy generating system, meteorological data for both wind and
solar is required. Evaluating the wind and solar energy characteristics together with their ability is the utmost important step for
economic viability of these renewable energy projects. Actually, the pdfs of wind speeds and solar irradiance constitute the wind
speeds and solar irradiance gathered over a long period. As a result, their information is important for evaluating the wind and solar
energy ability of a certain location.
A broad literature examination shows that several parametric probability distribution functions have been tested to predict both wind
speeds and solar irradiance, which are applied in reliability evaluation of these renewable resources. Probability distribution for
modelling global solar radiation of Ibadan, Nigeria was conducted by Ayodele and concluded that logistic distribution presents the
best probability distribution in modelling global solar radiation of Ibadan. In a similar fashion, four distribution methods were
examined and all four probability functions showed similar results for Hualien and Taitung by Tian Pau Chang in Taiwan. Many
other solar radiation probability distributions have been investigated around the different parts of the continent with Weibull
distribution function proving to be best distribution model for M’sila region, Algeria Razika and Nabila (2016), Akuffo and Brew-
2876