Forecasting net energy consumption of South Africa using Artificial Neural Network

Abstract

This work proposes the use of Artificial Neural Network (ANN) as a new approach to determine the future level of energy consumption in South Africa. Particle Swarm Optimization (PSO) was used in order to train Artificial Neural Networks. The population size, the percentage losses, the Gross Domestic Product (GDP), the percentage growth forecasts, the expected Final Consumption Expenditure of Households (FCEH) as well as the relevant manufacturing and mining indexes are the “drivers” values used for the forecasts. Three growth scenarios have been considered for the forecasting namely low, moderate and high (less energy intensive) scenarios. These inputs values for the period of 2014 to 2050, from the Council for Scientific and Industrial Research (CSIR), were used to test data and validate the use of this new approach for the prediction of electricity demand. An estimate of the annual electricity demand forecasts per scenario was calculated. Besides the speed of the computation, the proposed ANN approach provides a relatively good prediction of the energy demand within acceptable errors. ANN was found to be flexible enough, as a modelling tool, showing a high degree of accuracy for the prediction of electricity demand. It is expected that this study will contribute meaningfully to the development of highly applicable productive planning for energy policies.