Sequential Growth Model and NeuroAMI Neural Network for Distribution Transformer Load Forecasting

Abstract: Accurate distribution transformer load forecasting is essential for ensuring reliable power system operation, preventing transformer overloading, supporting asset management, and facilitating effective capacity expansion planning. However, increasing load demand, stochastic consumer behaviour, seasonal variations, and measurement uncertainties pose significant challenges to conventional forecasting techniques. This study presents a Sequential Growth Model integrated with a Neuronal Auditory Machine Intelligence (NeuroAMI) Neural Network for real-time distribution transformer load forecasting using historical electrical load time-series data. The proposed methodology incorporates a real-time discrete sampling model, finite-window normalization, data augmentation through noise injection, and trend- seasonal decomposition to improve data quality and model robustness. The NeuroAMI neural network employs auditory- inspired spectrogram feature extraction, heteroscedastic Gaussian negative log-likelihood optimization, regularized learning objectives, and online mini-batch gradient adaptation to accurately predict future transformer loading conditions. Historical transformer load data spanning 2008–2017 were utilized to forecast loading conditions from 2018–2027. The results demonstrated significant forecasting capability, with transformer load demand increasing from approximately 8,000 kW in 2017 to over 10,000 kW by 2027. The training process exhibited stable convergence, as the Negative Log- Likelihood loss decreased from approximately 80 to 20, while regularization loss reduced from about 15 to 5 over 100 training epochs. Furthermore, feeder-level forecasting revealed projected load growth from 1,500 kW to 1,850 kW for Elekahia feeder, 4,700 kW to 6,000 kW for Stadium Road feeder, and 3,600 kW to 4,300 kW for Rumukalagbor feeder. The study concludes that the proposed Sequential Growth–NeuroAMI framework provides an intelligent, adaptive, and reliable forecasting tool capable of supporting utility operational policies, preventive maintenance strategies, transformer capacity planning, and sustainable distribution network expansion.

Keywords: NeuroAMI, Transformer, Loads Forecast, Machine Learning