Accurate Probability Distribution Model Determination and Forecasting of Peak Load Demand in Nigeria
Ignatius K. Okakwu, Akintunde S. Alayande, Patience E. Orukpe |Pages: 186-194|

Abstract— This study aims at identifying the best-fit probability distribution and forecasting of the peak load demand in Nigeria. The data used was obtained from the National Control Centre (NCC), Oshogbo, Nigeria for a period of twenty years (1998-2017). Five different probability distribution functions and two forecasting models were used. The probability distribution functions explored include Normal, Log-normal, Gamma, Weibull and Logistic distribution from which the best was determined using two different goodness-of-fit. The two goodnessof-fit used are Akaike Information Criterion (AIC) and Schwartz Bayesian Criterion (SBC) while the two forecasting models include Auto Regression (AR) and Exponential Smoothing (ES). The best model is expected to have the lowest value for AIC, SBC, Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Theil Inequality Coefficient (TIC). The model that satisfies tests adequately was selected as the best fit. Results showed that the Log-normal distribution presents the best fitted distribution with AIC value of 327.5168 and SBC value of 329.5082 followed by Normal distribution with AIC of 327.5987 and SBC of 329.5902; Weibull distribution with AIC of 327.8540 and SBC of 329.8454; Gamma distribution with AIC of 328.0087 and SBC of 330.0002; and Logistic distribution of AIC of 328.3212 and SBC of 330.3127 respectively. The AR gave the best result among the two models with MAPE value of 0.21, MAE value of 12.55, RMSE of 173.40 and TIC value of 0.022. The results from this study will be very useful for decision makers, system operators, load forecasters, scheduling of electricity and potential investors in the power industry in Nigeria.