Software for predicting short-term generating capacity of solar power plants using artificial intelligence
Figure 1. Chairman Dr. Nguyen Quang Ninh
In this study, the research team reviewed methods for short-term forecasting of solar power plant generation capacity using artificial intelligence, presented in detail the methods, evaluated and selected appropriate methods for application to industrial-scale solar power plants in Vietnam. Next, the research team programmed and built software to forecast the short-term generation capacity of solar power plants in Vietnam. The result of the project is a forecasting method based on artificial intelligence techniques, to predict the value of reduced electricity output during the operation of the plant, thereby helping the plant owner create reasonable operating plans and maximum benefits in terms of reducing generation capacity.
Based on the advantages of the software model, in this study, the team used the software model to build a tool to forecast the output generation capacity of solar power plants in Vietnam. Not simply applying the software artificial intelligence algorithm, in the process of processing input data, to solve the problem that the actual output generation capacity of a solar power plant does not correspond to the possible capacity, due to power reduction requests from system operators, the authors proposed a new analysis technique to increase the properties of Long-Short Term Memory (LSTM), neural network to accurately predict output power (divide GHI range, add P/GHI coefficient and apply validation set). The model training process using validation data proposed by the authors is presented as shown below.
Figure 2. Diagram of training process using validation data and P/GHI coefficient in data processing
The forecast results verifying the generation capacity for solar power plants in actual operation under the following scenarios: one day of the rainy season, one day of the sunny season, one month of the rainy season, one month of the dry season, showing the forecasted capacity quite close to the actual generation capacity with an average MAPE absolute percentage error of less than 10%, meeting the proposed technical requirements, and the software's predictive performance is stable under experimental conditions.
Forecasting results show that the proposed new method is more effective than the old method, as the MAPE (Mean Absolute Percentage Error) forecast error is reduced by 6.059% and RMSE (Root Mean Square Error) is reduced by 6,710% (Figure 3).
Figure 3. Results of evaluating the effectiveness of the model using 10% validation compared to the model not using validation
The solution introduced above to forecast solar power output in the next 5 minutes is a step-by-step forecast. The output power forecast results of a solar power plant for this type of forecast can be input data for the Energy Management System (EMS) to provide real-time warning calculations for the system. The accuracy of these warnings is crucial in the operation of power systems powered by many solar renewable energy sources because it allows dispatchers to promptly take measures in emergency situations to limit the risk of frequency or voltage collapse in the system.
The results of the project can be applied in solar power plants in Vietnam. Forecasting software will help solar power plant operators predict the value of reduced electricity output during the plant's operation. From there, solar power plant owners can make reasonable operating plans and gain maximum benefits in terms of reducing generation capacity.
The research results of the project were published in 02 international articles in the SCIE category, Q1, 02 international articles in the Scopus category (Q3 and Q4). To be able to put the results of this research into practical application, the research team hopes to continue collecting data and calculating tests for solar power plants in Vietnam under different conditions and build a real-time operating forecast system for a solar power plant in Vietnam. From there, it serves as a basis for widely deploying this forecasting system at solar power plants across Vietnam.
Translated by Phuong Huyen
Link to Vietnamese version