Analysis of Electricity Consumption Data for Demand Forecasting in Abuja, FCT

Accurate electricity demand forecasting is critical for grid management and infrastructure planning in Nigeria's power sector, where chronic supply-demand imbalances persist. This study analysed monthly electricity consumption records for 14,500 residential and commercial consumers in Abuja Municipal Area Council, Federal Capital Territory, supplied by Abuja Electricity Distribution Company over 2018 to 2022. Data preprocessing handled missing meter readings and identified anomalous consumption spikes attributable to power restoration after outage events. Seasonal decomposition revealed a consistent peak in December and a secondary peak in April. Multiple linear regression, random forest, and LSTM neural network models were compared for demand forecasting performance. The LSTM model achieved the lowest RMSE of 12.4 megawatt-hours per month at the district aggregation level, outperforming regression by 21 percent. Random forest provided superior interpretability, with temperature and lagged demand as the most important predictors. Commercial consumers showed 1.6 times higher demand variability than residential consumers, complicating aggregate forecasts. Generator fuel cost was identified as a proxy variable for informal self-generation substitution behaviour. The study highlights the value of consumer-level disaggregation for improving grid planning. Recommendations include real-time smart metering deployment for Abuja consumers, integration of meteorological forecasts into AEDC dispatch planning tools, and a demand-side management incentive programme targeting peak period consumption reduction among large commercial users.

Keywords: demand forecasting, electricity consumption, LSTM neural network, Abuja AEDC, smart metering