A Comparative Study of Short-Term Load Forecasting in the Electricity Distribution Networks of Nigeria Using Deep Learning Algorithms

Abstract In the Nigerian distribution system, it is of essence that precise load forecasting be done to achieve an optimal utilization and reduce losses in the Nigerian power distribution network. Earlier load forecasting techniques that rely solely on statistical models like regression and ARIMA were used; however, these were not adequate to model the nonlinear behavior of electricity load as is found in Nigeria, where data quality issues and socioeconomic dynamics affect loads. Deep learning models including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), and transformer models are investigated and tested on Nigerian load data for short-term load forecasting. Each stage was implemented to handle data problems, including missing data and out-of-range values, and feature generation to incorporate calendar information and weather data and was evaluated on performance indicators such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Deep learning models performed far better than statistical models and generally ranked high in terms of accuracy, with the Transformer models being the best, followed by the LSTMs and GRUs, whereas CNN was moderately good but was not as effective in modeling the time dependencies in load forecasting as compared to the others. The models proved that there exists nighttime peak demand in Nigeria’s load patterns and that temperature is the most significant of all the weather attributes affecting Nigeria’s loads. The model findings prove that deep learning models are a viable alternative to the traditional methods, are scalable, and are suitable for use by Nigerian distribution companies. Deployment of GRU and LSTM could commence immediately, while a Transformer model can be deployed when data infrastructure in Nigeria’s distribution networks has sufficiently advanced.