Change the world

Centre of Expertise in Forecasting

Forecasting and Modelling Exhibits (FAME)

On June 3, 2011, COEF was not only launched but members also launched its first technical research outputs in form of Forecasting and Modelling Exhibits (FAME)

There exists a clear need among many sectors of society, including commerce, industry and Government, for some form of information regarding the future. Predictive models and statistical projection provide means of gaining a rational insight into the future. The Exhibits of the Centre plans to fill the existing gap in this area, by providing a service to external parties and stakeholders. In so doing, the FAME will provide the platform for post-graduate training of students, primarily from the Science Faculty, but including links with other faculties and Universities.

Below are abstracts from FAME. For more information contact the editor Mr. Abel Motsomi (details at the bottom of the page)

A NEW DISCRETE MODEL FOR PAIRED COMPARISONS

Morné Sjölander and Igor Litvine

This article introduces a new discrete paired comparisons model, the SL-model for paired comparisons, which has been designed for paired comparisons experiments in which objects' scores have a fixed predetermined upper limit. The underlying distribution of this model is a truncated negative binomial distribution. We show theoretically how this model can be used to make rank objects and make predictions and apply our model to rank the performance of tennis players.

In general it was found that investing in a diversified portfolio is better than investing in individual power sources.

STATISTICAL MODEL FOR EFFICIENT USE OF RENEWABLE ENERGY FROM DIFFERENT SOURCES

Edmund Ahame and Igor Litvine

The growth of the industry and population of South Africa urges to seek new sources of electric power, hence the need to look at alternative power sources. Power output from some renewable energy sources is highly volatile. %For instance power output from wind turbines or photovoltaic solar panels fluctuates between zero and the maximum rated power out.

To optimize the overall power output a model was designed to determine the best trade-off between production from two or more renewable energy sources putting emphasis on wind and solar. Different measures of volatility and risk – return plots were used to choose between the different investments options.

In general it was found that investing in a diversified portfolio is better than investing in individual power sources.

EVENTS DATABASE AND SHORT-TERM DEMAND FORECASTING

Thandekile Nyulu, Igor Litvine and Abel Motsomi

The most important criterion for choosing a forecast method is its accuracy, or how closely the forecast predicts the actual event. In short-term electricity demand forecast, hourly, daily and weekly consumption of electricity is the main objective. Short-term electricity demand is the most important load of the day by which electric businesses plans and conveys the loading of producing units in order to meet system demand. The accuracy of the conveying system depends on the accuracy of the forecasting algorithm used. The forecasting is the most vital instrument to predict the possible demand for the future. Regression analysis is one of the most important tools used for model analysis. The main objective of this project is to test the effect of South African holidays in electricity demand and discuss the daily disparity of electricity demand in South Africa.

FITTING OF STATISTICAL DISTRIBUTIONS TO WIND DATA IN PORT ELIZABETH

Kirshnee Moodley and Igor Litvine

Coastal South African cities like Port Elizabeth are said to possess a strong potential for wind energy. The probabilistic distributions of wind speeds and directions are essential for assessing the power potential in wind areas. These wind distributions are used for predicting the performance of wind energy conversion systems. Traditionally, the Weibull Distribution has been used to describe wind data. In this study, the Weibull distribution as well as alternative distributions which may also be used to represent wind speeds and directions are fitted to the data and compared. The objective of this study is to find out which distribution provides the most accurate and adequate fit for the wind data in Port Elizabeth.

CONSOLIDATION OF ENERGY DEMAND FORECASTS USING STATISTICAL TOOLS

Abel Motsomi and Igor Litvine

This paper has attempted to develop a consolidating instrument that will merge all the forecasts from different offices to one 'official forecast'. Such an instrument should be able to predict with accuracy the anticipated usage or demand. There are current shortfalls identified that lead to discrepancies between supply and demand by the electricity producer. Discrepancies (such as granularity of data) between the forecasters have created a need for an aggregating instrument. The current status quo shows several forecasters creating their own predictions, from different offices for different purposes. The results presented in this paper present a newly developed procedure of consolidating energy demand forecasts from different users and accounting for different time horizons.

A FORECASTING MODEL FOR PHOTOVOLTAIC MODULE ENERGY PRODUCTION

Paul Swanepoel and Igor N. Litvine and Ernest E. van Dyk

Energy is of concern for governments and economies all over the world. As conventional methods of energy production are facing the prospect of depleting fossil fuel reserves, economies are facing energy risks. With this tension, various threats arise in terms of energy supply security. A shift from intensive fossil fuel consumption to alternative energy consumption combined with the calculated use of fossil fuels needs to be implemented. Using the energy radiated from the sun and converted to electricity through photovoltaic energy conversion is one of the alternative and renewable sources to address the limited fossil fuel dilemma.

South Africa receives an abundance of sunlight irradiance, but limited knowledge of the implementation and possible energy yield of photovoltaic energy production in South Africa is available. Photovoltaic energy yield knowledge is vital in applications for farms, rural areas and remote transmitting devices where the construction of electricity grids are not cost effective. In this study various meteorological and energy parameters about photovoltaics were captured in Port Elizabeth (South Africa) and analyzed, with data being recorded every few seconds. A model for mean daily photovoltaic power output was developed and the relationships between the independent variables analyzed. A model was developed that can forecast mean daily photovoltaic power output using only temperature derived variables and time. The mean daily photovoltaic power model can then easily be used to forecast daily photovoltaic energy output using the number of sunlight seconds in a given day.

NEURAL NETWORKS FOR ELECTRICITY DEMAND FORECASTS

Mathys C. du Plessis and Christiaan J. Pretorius

Previous studies have shown that Neural Networks are very well suited to application in electricity usage forecasting. The fact that Neural Networks produced good results even when presented with noisy and non-linear data indicates that employing Neural Networks for such forecasting is worth considering, as opposed to more conventional statistical techniques which often cannot cope properly with noise and non-linearity. This paper describes specific Neural Networks that were created to perform electricity forecasts of various granularities. Several of the techniques employed by previous researches have been incorporated and a few novel approaches investigated. Experimental results are presented to prove that Neural Networks can produce reasonably accurate South African electricity usage forecasts.

Contact information
Mr Abel Motsomi
Contract Lecturer
Tel: +27 41 504 1392
abel.motsomi@mandela.ac.za