ENERGY CONSUMPTION FORECASTING IN SMART GRIDS USING DEEP LEARNING

Abstract

Smart Grids are intelligent electricity networks that are being adopted worldwide as modern grid systems have become complex. Networks with dynamic behaviour in Smart Grids require precise energy forecasting to achieve optimal management of loads while minimising operational expenses between various suppliers and consumers. This research develops a specific Al-based energy forecasting system that caters to Pakistan’s national energy network requirements. The system uses LSTM, GRU, Bidirectional LSTM, XGBoost, DeepAR together with historical data and socio economic variables to generate electricity demand predictions. Standard forecasting accuracy evaluation metrics like MAE, MSE, and R2 measure the model's performance. The system also implements a user-friendly web dashboard that utilises React and Supabase as well as Plotly as part of its modem technology foundation. The visual dashboard allows users to monitor predictions in addition to tracking patterns and automatically producing current reports. The developed system will help Pakistan's energy sector achieve smarter grid management by implementing data-driven decision systems which enhance power loss reduction and increase grid reliability.