Comparative Analysis of Neural Network Architectures for Short-Term Load Forecasting Using Real Operational Data: A Case Study of Al-Jufra Grid

Abstract

One of the most application in electric power system operation is the Short-term load forecasting (STLF), due to its role in supporting operational decisions and improving load management. In this paper, three different architectures of artificial neural networks (ANN) were studied and compared, which are Feed-Forward Neural Network (FFNN), Recurrent Neural Network (RNN), and Cascade Neural Network (CaNN), to evaluate their performance in short-term load forecasting. The study was applied using real operational data of Al-Jufra grid during the period from January to mid-April including temporal variables, historical load, and climate data inputs. The results showed that all models of ANN achieved good forecasting accuracy, with the RNN model outperforming the other models. The study also showed that data quality and the selection of appropriate inputs are important factors for improving model performance and operating electrical grids effectively.