Abstract

Technical losses significantly affect the efficiency and reliability of medium-voltage distribution networks in Nigeria. This study presents an ANN-based electro-thermal framework for analysing and predicting technical losses using 5,544 hourly operational data points from a distribution feeder modelled on the IEEE 33-bus system. Results indicate an average load of 49.97 MW with a peak of 103.30 MW, while technical losses average 2.26 MW and reach a maximum of 4.68 MW. The system exhibits a stable loss factor of 0.045 p.u (4.53%), confirming a strong proportional relationship between load and losses. Loss decomposition reveals a near-balanced contribution, with line losses accounting for 48.31% and transformer losses 43.89%, deviating from the conventional dominance of line losses. Electro-thermal analysis shows a relatively modest temperature influence, with loss variation ranging from −1.92% to 3.80% (average 0.84%). Approximately 18.58% of system operation occurs under high-loss conditions, where losses increase to an average of 3.69 MW. The ANN model achieves a mean absolute percentage error (MAPE) of 11.90% and R² of 0.9379, demonstrating good predictive capability. Annual energy loss is estimated at 12,550.38 MWh, corresponding to a cost of ₦1.255 billion. The proposed framework enhances loss visibility and supports data-driven monitoring, planning, and operational optimisation in distribution networks.

Key words: Technical losses; Artificial Neural Network (ANN); Electro-thermal modelling; Distribution systems; Loss prediction; Power system efficiency; Smart monitoring