Using hybrid algorithm to model and forecast electricity consumption: A case study of Uganda

Long term electricity consumption forecasting is used by decision makers to make decisions regarding system expansion planning. Over the past decade, research on electricity consumption forecasting has reported results as point forecasts. Specifically for long-term forecasting, point forecasts are of little interest because it is hard to use them to assess the financial risk associated with system expansion versus demand variability and forecasting uncertainty. In this study, firstly we use power, quadratic and Adaptive Neuro Fuzzy Inference System (ANFIS) forms to model Uganda's net electricity consumption using population, gross domestic product, number of subscribers and average electricity price as variables in the forecasting models. We optimize the parameters of power and quadrtaic forecasting models using a hybrid algorithm based on particle swarm optimization (PSO) and artificial bee colony (ABC) algorithms. The parameters of ANFIS model are optmized using particle swarm optimization and genetic algorithm. Secondly we model the forecast residuals using simple linear regression to obtain 90% significance level lower and upper error bounds. The lower and upper error bounds were used to construct predication intervals for Uganda's net electricity consumption up to year 2040. Finally we combine all the four models from the two methods to get a combined forecasting model. According to the combined forecast, in year 2040 Uganda's electricity consumption will be between [41,296 42,133] GWh indicating an annual average increase of 11.75%-10.64%