Short-Term Load Forecasting Using Multilayer Neural Networks: A Residential Complex Case Study

Short-term electric load forecasting (STLF) is a crucial tool for optimizing energy consumption and load management in power systems. This study aims to develop an accurate short-term electric load forecasting model using multilayer neural networks (MLNNs) that integrates historical electrical load data with meteorological variables. The dataset used for training the model was collected from a residential complex in Ahvaz, Iran, spanning from March 21, 2021, to August 24, 2021. It includes $\mathbf{7, 2 2 2}$ records of half-hourly electrical load data, along with the corresponding weather data such as temperature, humidity, wind speed, and atmospheric pressure. The primary objective of the model is to improve the prediction accuracy for energy consumption in dynamic environments, such as large buildings or industrial complexes, where various internal and external factors influence electricity demand. The results revealed that the model performed exceptionally well on the first day, with the forecasted loads closely aligning with the actual loads and minimal deviation. However, as the forecasting period extended, the accuracy gradually declined, especially in high-demand scenarios. Specifically, over the seven-day period, the forecasted loads were approximately $6.5 \%$ higher than the actual loads. Despite these discrepancies, the model effectively captured the overall trend of electricity demand, providing valuable insights for energy management and grid optimization. Overall, while the model’s performance can be enhanced, it provides a solid foundation for improving short-term electric load forecasting and serves as a valuable tool for demand-side management and energy efficiency strategies.