Related Experiment Video
Updated: Oct 12, 2025
10:46
A Method of Trigonometric Modelling of Seasonal Variation Demonstrated with Multiple Sclerosis Relapse Data
Published on: December 9, 2015
10.8K
An ensemble method to forecast 24-h ahead solar irradiance using wavelet decomposition and BiLSTM deep learning
Pardeep Singla1, Manoj Duhan1, Sumit Saroha2
1Deenbandhu Chhotu Ram University of Science & Technology, Sonepat, India.
Earth Science Informatics
|November 22, 2021
Summary
This study introduces an ensemble model combining wavelet transform and bidirectional long short-term memory networks for accurate 24-hour solar global horizontal irradiance (GHI) forecasting. The proposed model significantly outperforms existing methods, improving grid management and renewable energy integration.
Area of Science:
- Renewable Energy Systems
- Artificial Intelligence in Energy
- Time Series Forecasting
Background:
- Exponential growth in solar power necessitates accurate solar global horizontal irradiance (GHI) prediction for grid management.
- The stochastic nature of solar power presents significant challenges for precise GHI forecasting.
- Accurate GHI forecasting is vital for effective grid planning, scheduling, and balancing.
Purpose of the Study:
- To propose an advanced ensemble model for accurate 24-hour ahead solar GHI forecasting.
- To enhance forecasting accuracy by integrating wavelet transform (WT) with bidirectional long short-term memory (BiLSTM) deep learning.
- To evaluate the proposed model's performance against benchmark and other state-of-the-art forecasting methods.
Main Methods:
- Decomposition of solar GHI time series data into intrinsic model functions (IMFs) using wavelet transform (WT).
- Reduction of IMF series by combining wavelet decomposed components (D1-D6) based on experimental analysis.
- Application of trained standalone BiLSTM networks to each IMF sub-series for individual forecasting.
- Reconstruction of forecasted sub-series values to generate the final solar GHI forecast.
Main Results:
- The proposed WT-BiLSTM ensemble model significantly outperformed benchmark (naïve predictor), standalone LSTM, GRU, and BiLSTM models.
- The model achieved substantial reductions in monthly average Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE).
- Achieved a coefficient of determination (R²) of 0.94 and a forecast skill (FS) of 47% compared to the benchmark.
Conclusions:
- The proposed ensemble model offers a superior approach for accurate and reliable 24-hour ahead solar GHI forecasting.
- Integration of WT and BiLSTM effectively captures complex patterns in solar irradiance data.
- The model's enhanced accuracy supports improved grid management and facilitates higher penetration of solar energy.
