Optimizing LSTM S2S models with Evolutionary Mating Algorithm (EMA) for direct multi-step forecasting of household electrical power consumption
Accurate prediction of household electrical power usage is vital for efficient distribution, cost savings, and sustainability. Traditional models often struggle with complex, shortterm, multi-step time series forecasting. This study proposes a hybrid approach that combines a Long Short-Term Memory S...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | en |
| Published: |
IEEE
2026
|
| Subjects: | |
| Online Access: | https://umpir.ump.edu.my/id/eprint/47465/1/EMAPlus-optimized%20adaptive%20convergence%20prescribed%20performance%20control.pdf https://umpir.ump.edu.my/id/eprint/47465/ https://doi.org/10.1109/ICSECS65227.2025.11279218 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate prediction of household electrical power usage is vital for efficient distribution, cost savings, and sustainability. Traditional models often struggle with complex, shortterm, multi-step time series forecasting. This study proposes a hybrid approach that combines a Long Short-Term Memory Sequence to Sequence (LSTM S2S) model with the Evolutionary Mating Algorithm (EMA) to optimize the model settings. The proposed model is tested on a public dataset of household power using seven input features and 48 hours of past data to predict the next 24 hours of power consumption using a direct multi-step approach.Additionally, a comparative analysis was made with standard LSTM, Gated Recurrent Units (GRU), and Recurrent Neural Networks (RNN). The result showed that LSTM S2S with EMA achieved the best performance, with a lower RMSE of 0.110, outperforming RNN (0.282), GRU (0.340), and standard LSTM (0.321). EMA successfully identifies optimal parameters, and EMA-optimized model achieved a significant improvement in accuracy over to baseline methods. This study highlights potential for smart grid and energy management applications. However, it is limited to a single household and does not reflect differences in household types, locations, or usage patterns. Future work will expand the scope using diverse datasets, longer forecast horizons, and integration of renewable energy sources for more practical and scalable solutions. |
|---|