Speaker
Description
With the rising awareness of green manufacturing, the importance of in- volving green energy into the manufacturing process has become a crucial is- sue. However, sustainable energy such as solar power, requires a reliable pre- dicting model for maximizing the power gathered from the solar power system. This research presented a novel machine learning method for predicting the solar radiation in a certain area, which is crucial for areas like Tai- wan to develop a stable, sustainable energy source for manufacturing. The data presented in this work is open-sourced from NASA which gathered important indexes for predicting solar radiation including, temperature, humidity, wind conditions, and the time of sunrise and sunset.
In this research, several machine learning methods are compared. Using the data of the previous five data points to predict the next time instance data. The results showed that the transformer model has the highest prediction results, gaining a result of RMSE 85.17, MAE 42.01, and R2 0.93. Two optimization algorithms, Quasi-Monte Carlo (QMC) and Tree-Structued Parzen Estimator (TPE), are implemented to opti- mized the hyper-parameters including dropout rate and the number of encoder layers of the transformer model. In a normal use case, during the training process, hyperparameters are tuned with trial-and-error method. The final results showed that through optimization the model gained better prediction results. The QMC method improved the mod- els’ performance with a result of RMSE 77.96, MAE 32.18 and R2 0.94.
The overall results demonstrate the performance potential of a well- optimized transformer model in predicting solar radiation. This research may be the foundation for research interests such as energy manage- ment, which should also benefit in green manufacturing. In this re- search, a novel method combining QMC into the tuning of the hyperpa- rameters are also crucial for future implementation of machine learning methods.
