Electricity load forecasting is essential for ensuring steady power supply and planning. In Nigeria, electricity distribution companies face persistent challenges of transformer overloading, poor voltage regulation, and frequent power interruptions, largely due to inadequate demand prediction. This research applies deep learning models to forecast short-term electricity load demand using six months of historical data obtained from an electricity distribution company. Five deep learning models viz: Bi Gated Recurrent Unit (Bi GRU), hybrid of Convolutional Neural Network and Gated Recurrent Unit (CNN-GRU), Temporal Convolutional Network (TCN), Long-Short Term Memory (LSTM), and Gated Recurrent Unit (GRU) were trained and evaluated using Mean Absolute Error (MAE), Root Mean Square Error, and coefficient of determination (R²) metrics. Among them, the GRU model achieved the best performance with an average MAE of 0.0292, RMSE of 0.0514, and R² of 0.5237. The trained model was used to forecast the next 24 hours loads, each feeder was mapped with their corresponding transformers of various stations, using the ratings of each to estimate transformer loading percentages based on the Nigerian Electricity Regulatory Commission (NERC) standard power factor of at least 0.85. Results revealed, among other transformers Gumel the TR1 was overloaded from 6 AM of 26/02/2024 to 4 AM of 27/02/2024, it has a rating of 7.5MVA, the predicted load at 6 AM was 5.7875MW which gives 90.79 overloaded, indicating potential overload risks if corrective measures are not taken. The study demonstrates the effectiveness of GRU-based forecasting in predicting short-term load demand and provides an insight to improve transformer efficiency and highlight potential overload for better distribution and system management.