THE DEVELOPMENT OF SOLAR POWER PREDICTION MODEL FOR PERFORMANCE EVALUATION USING EVENT-TRIGGERED UPDATE METHOD
The utilization of solar photovoltaic power plants (PLTS) is becoming increasingly popular due to the availability of unlimited energy sources. However, the performance of PLTS is highly dependent on fluctuating weather conditions and solar panel degradation. Therefore, a model is needed to accurate...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/74469 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | The utilization of solar photovoltaic power plants (PLTS) is becoming increasingly popular due to the availability of unlimited energy sources. However, the performance of PLTS is highly dependent on fluctuating weather conditions and solar panel degradation. Therefore, a model is needed to accurately represent the physical system in real-time and predict the power output of PLTS. Presently, the PLTS power prediction model used in the Energy Management Laboratory has not been updated since June 2022 and produces inaccurate predictions.
In this research, we aim to develop the existing prediction model and conduct periodic evaluations to maintain its efficiency and performance.
The model development is accomplished by optimization of the deep neural network (DNN) architecture by tuning hyperparameters to achieve an optimal model. The optimized model is then used for performance evaluation using the event-triggered update (ETU) method. This method updates the prediction model if the evaluation results exceed the tolerance limits for 4th consecutive days. The tolerance limits are a mean absolute error (MAE) of 160 W and a root mean square error (RMSE) of 280 W.
During the optimization process, adjustments are made to the number of neurons, activation functions, optimization functions, epochs, batch size, and learning rate. The optimization results show a significant decrease in MAE by 83.64% and RMSE by 78.77%, with an MAE value of 78 W and an RMSE value of 173 W. The optimized model is implemented on a digital twin (DT) platform, and its performance is evaluated from January 1, 2023, to June 14, 2023. The evaluation results indicate that no updates are required for the prediction model used, demonstrating the ability of the optimized model to adapt to fluctuating weather conditions.
|
|---|