|Project's title||Research on methodology and building software to forecast the short-term output power of solar power plants using artificial intelligence|
|Research hosting institution||Institute of Energy Science|
|Project leader’s name||Dr. Nguyen Quang Ninh|
|Project duration||01/01/2021 - 31/12/2022|
|Project’s budget||600 million VND|
|Goal and objectives of the project||
Research on methodology and building software to forecast the short-term output power of solar power plants using artificial intelligence
- Theoretical results: In this study, the short-term forecasting methods of the solar power plant's output power using artificial intelligence have been reviewed, presented details, and selected the suitable methodology to apply for an industrial-scale solar power plant in Vietnam. Based on the Long Short-Term Memory method, LSTM, the authors have proposed a solution to process input data in cases the solar power plant operates in curtailment conditions, and proposed a new training algorithm to archive better prediction results. Then a forecasting software based on the proposed method has been built, and some calculating experiments according to the proposed scenarios: one day in the rainy season, one day in the dry season, one month in the rainy season, and one month in the dry season have been carried out for a solar power plant in Vietnam. The results show that the predicted power is quite close to the actual generated power with the Mean Absolute Percentage Error, MAPE, below 10%, meeting the technical requirements, and the software works stably under experimental conditions.
|Novelty and actuality and scientific meaningfulness of the results||
The main purpose of this study is not only to suggest the direction of selecting suitable features and structure of the LSTM model but also to create a feasible procedure to forecast the short-term output power of a solar power plant in Vietnam. In this study, the authors focus on advanced training techniques for LSTM neural networks by proposing the use of validation sets and the application of GHI interval division techniques, adding the P/GHI factor in the process of processing data collected initially under the curtailment conditions. The results show that the proposed new method is more efficient than the old method with both MAPE (Mean Absolute Mean Percentage Error) and RMSE (Original Mean Squared Error) forecasting errors.
|Products of the project|| |
- Scientific papers in referred journals (list):
|Research area|| |
The results of the project can be applied to solar power plants in Vietnam. Forecasting software will help solar power plant operators predict the value of output power reduction during the plant's operation. Based on that, solar power plant owners can make a reasonable operation plan and get maximum benefits in curtailment conditions.
|Images of project|| |