DEVELOPMENT OF LONG SHORT-TERM MEMORY –RECURRENT NEURAL NETWORK MODEL FOR THE ASSESSMENT AND PREDICTION OF VOLTAGE SAG IN ADO-EKITI SECONDARY DISTRIBUTION NETWORK

Abstract

Voltage sag is a significant power quality disturbance that adversely affects the reliable operation of sensitive equipment and overall industrial productivity in secondary distribution networks. In the Ado-Ekiti metropolis, Nigeria, recurring voltage instability driven by load fluctuations, feeder overloading, and network topology limitations necessitates the adoption of advanced, data-driven monitoring and forecasting approaches. This study presents a predictive framework based on a Long Short-Term Memory–Recurrent Neural Network (LSTM-RNN) architecture designed to capture nonlinear temporal dependencies associated with voltage variations. The model was developed and validated using high-resolution historical measurements obtained from six distribution feeders. To enhance generalization and reduce overfitting, regularization techniques including L2 weight decay, dropout, and early stopping were incorporated during training. Model performance was evaluated using standard statistical indicators, including Root Mean Square Error (RMSE) and the coefficient of determination R², alongside operational criteria for disturbance detection defined by voltage sag conditions below 0.8p.u. The results indicate strong predictive performance, with R² values ranging from 0.83 to 0.93 across the studied feeders.In particular, the model achieved an R² value of 0.89 and a Mean Absolute Error (MAE) of 1.53 V for the Adebayo feeder, demonstrating effective tracking of seasonal voltage variations and recurring instability patterns.These findings confirm that the proposed LSTM-RNN framework provides a reliable and scalable approach for power quality assessment in low-voltage distribution systems, supporting improved operational planning and enhanced network reliability in developing power infrastructure.