What question did this study set out to answer?

The study aims to improve forecasting accuracy for power loads in the context of increasing renewable energy integration.

April 22, 2026Open Access

A multi-component hybrid model for renewable energy integrated grid power load forecasting: DWT, SAE, and FA-LSTM

Key Points

The study aims to improve forecasting accuracy for power loads in the context of increasing renewable energy integration.
Applied discrete wavelet transform (DWT) for multi-scale decomposition of power load data.
Used stacked autoencoder (SAE) for denoising decomposed high-frequency components.
Optimized long short-term memory (LSTM) network parameters with firefly algorithm (FA).
Achieved mean absolute percentage errors (MAPE) of 2.540%, 12.90%, and 26.110% for Fujian, Xinjiang, and Gansu respectively.
Current approach outperformed comparison models which had higher MAPE values.
Significantly enhanced accuracy and robustness of power load forecasting across different regions.

Abstract

As renewable penetration deepens and grid operating conditions grow more intricate, load data exhibit significant non-stationarity and multi-scale characteristics, leading to certain limitations in modeling and forecasting accuracy of traditional methods. To tackle this issue, this research suggests a power load forecasting (PLF) approach that utilizes the discrete wavelet transform (DWT), stacked autoencoder (SAE), and long short-term memory network (LSTM) optimized by firefly algorithm (FA). Firstly, DWT is used to perform multi-scale decomposition on the raw power load series, separating the signal into a high-frequency residual and a low-frequency trend. Then, SAE is applied to denoise the decomposed high-frequency components, effectively filtering out residual noise and improving data quality. Finally, FA is employed to optimize the key parameters of LSTM, achieving accurate modeling of features in different frequency bands. Empirical results indicate that the suggested approach reaches mean absolute percentage errors (MAPE) of 2.540%, 12.90%, and 26.110% in forecasting for Fujian, Xinjiang, and Gansu regions, respectively, while the best MAPE of other comparison models is 4.370%, 43.270%, and 37.330%. This significantly improves the accuracy and robustness of PLF in different regions, providing a new idea and method for power data forecasting. Code is available at https://github.com/Siri-scut/DWT-SAE-FA-LSTM.git.

Bookmark

View Full Paper