Time series analysis of AMI data and comparative energy demand forecasting using deep learning models in a smart grid scenario

Abstract

The dynamic variations of consumer load demand across different temporal scales—short (hourly), medium (daily), and long-term (monthly to yearly) and the increased integration of renewable energy sources have significantly enhanced the complexity of modern smart power grids. Accurate prediction of energy demand is important for grid stability optimization, optimal resource optimization, and planning operations. This study utilizes the American Electric Power Company (AEP) data set of 121,273 hourly electricity consumption records (in megawatts) to develop forecast models solely based on time series data. Deep learning methods were used to improve forecasting accuracy across different timescales. Short-term forecasts used Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Temporal Convolutional Networks (TCN), and Sequence-to-Sequence (Seq2Seq) architectures, while medium-term forecasting used Transformer, CNN-LSTM, and Recurrent Neural Networks (RNNs). In long-term forecasting, Prophet, Feedforward Neural Networks (FNN), and Neural Basis Expansion Analysis for Time Series (N-BEATS) were used. Models were validated with standard performance metrics, such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (RMSE), alongside graphical analysis that proved forecasting accuracy. Results revealed that the Seq2Seq model attained maximum precision for short-term forecasting with the minimum RMSE measure value of 0.013008, while RNN exhibited improved performance in medium-term forecasting with an RMSE of 766.7151. In long-term forecasting, the N-BEATS model attained the best performance with an RMSE of 0.122733. These results prove the capability of deep learning models in improving energy demand forecasting, enabling smart resource allocation, enhanced load balancing, and efficient demand-side management, ultimately leading to the development of intelligent and sustainable power grids.