Development of a Load Forecasting Model for the Benin Electricity Distribution Company Service Area

This project develops a short-term electrical load forecasting model for the service area of the Benin Electricity Distribution Company in Edo and Delta States, South South Nigeria. Accurate short-term load forecasting is essential for optimal scheduling of generation and network switching decisions, yet BEDC currently lacks an in-house forecasting capability, relying on historical average demand estimates that do not account for seasonal, weather, and calendar-driven load variations. Historical hourly load data for 12 months were obtained from BEDC and supplemented with temperature and humidity records from the Nigerian Meteorological Agency for Benin City. The data were preprocessed to remove meter outage periods and outlier readings before model training. A multiple linear regression model and an artificial neural network model were both developed and compared. The neural network architecture comprises an input layer receiving load, temperature, hour of day, and day of week variables, two hidden layers of 15 and 10 neurons respectively, and a single output neuron predicting the next hour's load. Training used 70 percent of the dataset, with 15 percent each for validation and testing. The neural network model achieves a mean absolute percentage error of 3.8 percent on the test set, outperforming the regression model at 7.1 percent. Seasonal decomposition analysis confirms a pronounced evening peak between 7 pm and 10 pm throughout the year. Recommendations include integrating the model into BEDC's control room software and retraining quarterly with updated load data. Keywords: load forecasting, artificial neural network, electricity distribution, Benin, demand prediction.