loading page

Enlarging Forecast Horizon for Residential Load Prediction using Sequence-To-Sequence LSTM
  • Abhishu Oza,
  • Dhaval K Patel
Abhishu Oza
School of Engineering and Applied Science, Ahmedabad University

Corresponding Author:[email protected]

Author Profile
Dhaval K Patel
School of Engineering and Applied Science, Ahmedabad University


The power system is undergoing a large change towards renewable energy technologies. While using these energy sources, managing the generation, storage and distribution of energy can be optimized with information about future energy consumption. The forecasting of consumption load for individual residents plays a key role for load balancing but is a challenging task due to the volatile nature of individual consumption. Due to this reason, current literature has only been limited to forecasting individual load to a small window in the future. In this paper, we introduce a Sequence-To-Sequence Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) framework to generate a 24 hour load forecast. We show comparisons with other deep neural network models of 1) Model performance over varying forecast window sizes, 2) Average model performance over multiple houses and 3) Performance for forecasting the aggregated load of all houses. We also conduct analyses on the forecasts to show performance improvement for households with consistent load patterns and to detect model degradation. Our extensive experiments show that the Sequence-To-Sequence LSTM RNN can significantly increase the forecast window and performs best for all scenarios.
12 Feb 2024Submitted to TechRxiv
14 Feb 2024Published in TechRxiv