Multi-level Stacking of Long Short Term Memory Recurrent Models for Time Series Forecasting of Solar Radiation
- Resource Type
- Conference
- Authors
- Al-Hajj, Rami; Assi, Ali; Fouad, Mohamad M.
- Source
- 2021 10th International Conference on Renewable Energy Research and Application (ICRERA) Renewable Energy Research and Application (ICRERA), 2021 10th International Conference on. :71-76 Sep, 2021
- Subject
- Bioengineering
Components, Circuits, Devices and Systems
Power, Energy and Industry Applications
Training
Stacking
Time series analysis
Solar energy
Predictive models
Data models
Stability analysis
LSTM
Gated Models
Recurrent Neural Models
Solar Radiation
Multi-Level Stacking
Energy Forecasting
- Language
- ISSN
- 2572-6013
The ability of predicting solar radiation strength is particularly important for the best administration of power grids involving photovoltaic solar energy. The production-consumption balance of energy is one of the most important characteristics that should be maintained in energy grids. A qualified solar energy short-term can help in mitigating such type of problems, and therefore helps in improving the reliability of the overall energy system. The contribution of the present research consists of proposing and validating multi-stacking model that consists of a committee of recurrent long short-term memory (LSTM) models to predict the strength of solar radiation one day ahead. The individual models of the committee have been generated by modifying the internal structure and parameters of a base recurrent LSTM model. A multi-level structure that consists of two layers of stacking meta-models have been examined and validated through a cross validation strategy. The proposed models have been trained on time series dataset of solar radiation records, and then validated and tested with on subset of data records over one year using correlation and statistical metrics.