ANALYSIS OF ELECTRIC POWER GENERATION IN NIGERIA

Abstract

ABSTRACT Electric power generation in Nigeria is grossly inadequate, leading to high electricity supply-demand imbalance and consequently, low productivity. This study analysed the Nigerian electric power generation from 2004 to 2020. A statistical model was developed, evaluated and used to forecast monthly power generation in Nigeria from 2021 to 2030. In carrying out this research, a 17- year monthly generated power (2004-2020) data were collected from the National Control Centre, Osogbo, Nigeria while the corresponding weather data which include temperature, relative humidity, evaporation, rainfall and sunshine were obtained from the Nigeria Meteorological Agency, Abuja, Nigeria. The time series power generation dataset was tested for stationarity using the Augmented Dickey-Fuller (ADF) test. The dataset was differenced for ADF statistics higher than the critical values and the P-value of 0.05 to remove trend, seasonality and outliers that account for its non-stationarity to improve the model accuracy. Six stochastic time series models comprising seasonal autoregressive integrated moving average (SARIMA), seasonal autoregressive integrated moving average with exogenous variables (SARIMAX), Prophet, exponential smoothing state space (ETS), long short-term memory (LSTM) and hybrid SARIMA-LSTM were developed and evaluated. Eighty percent of the 2004-2019 data was used for model training while twenty percent of 2019-2020 data was used for testing. The accuracy of the models was assessed using three metrics which are mean absolute percentage error (MAPE), mean square error (MSE) and coefficient of determination (R²) score. Python software tool was used to further authenticate the models. Parameters of the models were estimated using maximum likelihood method. The results of the model evaluation and diagnostic checks showed that SARIMA (0, 1, 1) (0, 1, 1) (12) model fit yielded the lowest MAPE of 20.84% and MSE of 17, 975.47; indicating that it best captured the trend and seasonality of the dataset. An R² score of 0.2259 for the SARIMA model, while low, is still positive and acceptable. The SARIMAX model with MAPE, MSE and R2 score of 26.56%, 24,346.47 and -0.4200, respectively, performed less effectively compared to SARIMA model, despite the addition of exogenous variables. This suggests that the exogenous variables might not significantly enhance the power generation forecast or that a more optimised configuration is needed. The ETS, LSTM, Prophet and hybrid SARIMA-LSTM models had MAPE of 25.52, 33.31, 25.11 and 56.01%; MSE of 336,553.46, 460,134,500.24, 305,422,656.96 and 679,097,189.60; R² scores of -0.0087, -0.1872, -0.0850 and 0.0562, respectively. The relatively high error in ETS, Prophet, LSTM and SARIMA-LSTM models indicates that they could not handle the data's monthly seasonality and trends. The SARIMA (0,1 1) (0,1,1) (12) model, therefore, proved useful for forecasting future power needs of Nigeria, having performed best across all metrics and capturing the linear trend and seasonality effectively. The forecasted power generation ranged from 100,045.222 MW in January 1, 2021 to 124,354.978 MW in December 1, 2030. This study has shown that the developed SARIMA model can be appropriately deployed in analysing the electric power needs of Nigeria for strategic energy planning and development of power generation infrastructure.