Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart Grids

Accurate load forecasting is essential for ensuring the stability and efficiency of modern power systems, particularly in the context of increasing renewable energy integration. This study proposes an advanced forecasting approach using Recurrent Neural Networks (RNN) with Long Short-Term Memory (LS...

Full description

Saved in:
Bibliographic Details
Main Authors: Muhammad Asghar Majeed, Sotdhipong Phichaisawat, Furqan Asghar, Umair Hussan
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11021601/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850221991684997120
author Muhammad Asghar Majeed
Sotdhipong Phichaisawat
Furqan Asghar
Umair Hussan
author_facet Muhammad Asghar Majeed
Sotdhipong Phichaisawat
Furqan Asghar
Umair Hussan
author_sort Muhammad Asghar Majeed
collection DOAJ
description Accurate load forecasting is essential for ensuring the stability and efficiency of modern power systems, particularly in the context of increasing renewable energy integration. This study proposes an advanced forecasting approach using Recurrent Neural Networks (RNN) with Long Short-Term Memory (LSTM) to enhance electricity demand prediction. The research analyzes key temporal patterns in historical load, renewable energy generation, and weather data to improve forecasting precision. RNN-based LSTM, with its gated memory structure, effectively captures long-term dependencies and mitigates vanishing gradient issues, making it highly suitable for modeling complex time-series data in power systems. Its ability to retain crucial past information across extended sequences enables precise load forecasting, even in the presence of fluctuations and intermittent renewable energy sources. To validate the effectiveness of the proposed technique, the model was implemented and tested using MATLAB/Simulink, ensuring realistic simulation of power system dynamics. Through a comprehensive comparative analysis, RNN-LSTM is found to exhibit superior predictive accuracy, achieving the lowest RMSE (2.2889), MAE (1.1041), and MAPE (1.538%), while also demonstrating faster convergence, lower inference time, and higher stability. Furthermore, LSTM demonstrates superior convergence speed, lower inference time (0.00098 s), and high stability (0.9091), which confirms its robustness. Compared to other deep learning techniques such as GRU, standard RNN, and CNN, RNN-LSTM outperforms in capturing long-range dependencies and learning from temporal patterns effectively. The convergence analysis further confirms the robustness of RNN-LSTM, making it the most suitable model for dynamic and complex power system environments. These findings underscore the significance of RNN-based LSTM in enhancing load forecasting reliability, ultimately contributing to improved grid stability, energy management, and operational efficiency.
format Article
id doaj-art-2a1044f73fc543dfaf1df8ea3aea61bc
institution OA Journals
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-2a1044f73fc543dfaf1df8ea3aea61bc2025-08-20T02:06:31ZengIEEEIEEE Access2169-35362025-01-0113990189903110.1109/ACCESS.2025.357630311021601Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart GridsMuhammad Asghar Majeed0https://orcid.org/0009-0007-1034-8413Sotdhipong Phichaisawat1https://orcid.org/0000-0002-6300-2914Furqan Asghar2https://orcid.org/0000-0002-6159-2236Umair Hussan3https://orcid.org/0000-0001-6302-3582Department of Electrical Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok, ThailandDepartment of Electrical Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok, ThailandDepartment of Energy Systems Engineering, Faculty of Agricultural Engineering and Technology, University of Agriculture, Faisalabad, PakistanCollege of Mechatronics and Control Engineering, Shenzhen University, Shenzhen, ChinaAccurate load forecasting is essential for ensuring the stability and efficiency of modern power systems, particularly in the context of increasing renewable energy integration. This study proposes an advanced forecasting approach using Recurrent Neural Networks (RNN) with Long Short-Term Memory (LSTM) to enhance electricity demand prediction. The research analyzes key temporal patterns in historical load, renewable energy generation, and weather data to improve forecasting precision. RNN-based LSTM, with its gated memory structure, effectively captures long-term dependencies and mitigates vanishing gradient issues, making it highly suitable for modeling complex time-series data in power systems. Its ability to retain crucial past information across extended sequences enables precise load forecasting, even in the presence of fluctuations and intermittent renewable energy sources. To validate the effectiveness of the proposed technique, the model was implemented and tested using MATLAB/Simulink, ensuring realistic simulation of power system dynamics. Through a comprehensive comparative analysis, RNN-LSTM is found to exhibit superior predictive accuracy, achieving the lowest RMSE (2.2889), MAE (1.1041), and MAPE (1.538%), while also demonstrating faster convergence, lower inference time, and higher stability. Furthermore, LSTM demonstrates superior convergence speed, lower inference time (0.00098 s), and high stability (0.9091), which confirms its robustness. Compared to other deep learning techniques such as GRU, standard RNN, and CNN, RNN-LSTM outperforms in capturing long-range dependencies and learning from temporal patterns effectively. The convergence analysis further confirms the robustness of RNN-LSTM, making it the most suitable model for dynamic and complex power system environments. These findings underscore the significance of RNN-based LSTM in enhancing load forecasting reliability, ultimately contributing to improved grid stability, energy management, and operational efficiency.https://ieeexplore.ieee.org/document/11021601/Recurrent neural networksload forecastingtemporal dependency modelinglong short-term memorygated recurrent unitsconvolutional neural networks
spellingShingle Muhammad Asghar Majeed
Sotdhipong Phichaisawat
Furqan Asghar
Umair Hussan
Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart Grids
IEEE Access
Recurrent neural networks
load forecasting
temporal dependency modeling
long short-term memory
gated recurrent units
convolutional neural networks
title Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart Grids
title_full Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart Grids
title_fullStr Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart Grids
title_full_unstemmed Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart Grids
title_short Data-Driven Optimized Load Forecasting: An LSTM-Based RNN Approach for Smart Grids
title_sort data driven optimized load forecasting an lstm based rnn approach for smart grids
topic Recurrent neural networks
load forecasting
temporal dependency modeling
long short-term memory
gated recurrent units
convolutional neural networks
url https://ieeexplore.ieee.org/document/11021601/
work_keys_str_mv AT muhammadasgharmajeed datadrivenoptimizedloadforecastinganlstmbasedrnnapproachforsmartgrids
AT sotdhipongphichaisawat datadrivenoptimizedloadforecastinganlstmbasedrnnapproachforsmartgrids
AT furqanasghar datadrivenoptimizedloadforecastinganlstmbasedrnnapproachforsmartgrids
AT umairhussan datadrivenoptimizedloadforecastinganlstmbasedrnnapproachforsmartgrids