| dc.description.abstract | According to Taiwan′s 2050 net-zero emission path and strategy, it is projected that renewable energy power generation will comprise up to 60% of the total energy generation. Solar photovoltaics have experienced the most rapid growth among all renewable energy sources in recent years, accounting for 60% of total renewable energy generation in 2020. In order to effectively monitor short-term changes in solar power generation, it is crucial to accurately understand the dynamics of clouds. Clouds exhibit significant temporal and spatial variability, including factors such as cloud cover, cloud type, cloud thickness, cloud altitude, cloud structure, and the relative position of clouds in relation to the sun. Therefore, a comprehensive understanding of these cloud characteristics is essential for accurately predicting solar power generation trends. The solar energy field can experience significant fluctuations in solar radiation within a short timeframe, leading to implications for power generation efficiency, power dispatch, and grid stability. In this research, the calculation of cloud cover and radiation will be conducted using image processing and the red-blue threshold method. The data will be obtained from the all-sky camera that is currently installed in the Chiayi Weather Station. Additionally, the study aims to predict short-term changes in cloud cover and radiation. To validate the results, high-resolution radiation observation data from the Chiayi Station of Taiwan′s Baseline Surface Radiation Network (BSRN) will be utilized for comparison and verification purposes.
The findings indicate that the utilization of Aerosol Optical Depth (AOD) in the cloud cover algorithm allows for the assessment of atmospheric gray masking. By applying various thresholds, a more precise calculation of cloud cover can be achieved compared to the original cloud cover products in the presence of high pollution gray imagery. The radiation model developed in this study demonstrates a strong agreement with the radiation observation data obtained from BSRN. The coefficient of determination (R2) is calculated to be 0.94, indicating a high level of consistency. Additionally, the root-mean-square error (RMSE) is found to be only 55.8W m-2, further supporting the accuracy of the model. The observed data and the model results exhibit a difference of 93.1%, highlighting the model′s ability to effectively capture rapid changes in radiation within a short time period. The findings also indicate that by computing radiation levels in various scenarios, it is possible to differentiate scenarios with high transient radiation values. The present study incorporates a short-term prediction method for radiation levels, which is based on the changing trends of model parameters. The prediction R2, within a ten-minute timeframe, exceeds 0.81, and the majority of the RMSE are below 100 W m-2. Furthermore, the accuracy of the predictions reaches 80.3% within a 10-minute interval. These findings demonstrate that the established prediction method effectively captures rapid short-term fluctuations in radiation levels. In addition to its potential application in other stations equipped with all-sky cameras in the future, this mode can also be executed with limited computing resources. As a result, it can be utilized with a compact computer and an all-sky camera module to observe and calculate the actual solar energy field. This capability offers a valuable reference for monitoring the fluctuation patterns of solar energy-related companies within a brief timeframe. | en_US |