| 摘要: | 依照臺灣2050淨零排放路徑及策略，預計再生能源發電占總發電的比例將高達60~70%，而近年來太陽光電的成長也是所有再生能源中最快速的，在2020年時已占所有再生能源總量的六成，若要掌握短時太陽能的發電變化，必須精準掌握雲量與雲的動向，其中雲有很大的時空變異性，如雲量、雲屬、雲厚、雲高、雲的結構及雲與太陽的相對位置等，皆容易使太陽輻射短時內有較大起伏，影響太陽能案場發電效率、電力調度及電網穩定性。本研究將透過目前嘉義氣象站內所架設使用的全天空照相機，透過影像處理與紅藍閾值方法計算出雲量與輻射量，並進一步預測短期雲量及輻射的變化情形，也使用台灣標準地面輻射觀測網(Baseline Surface Radiation Network，BSRN)嘉義站之高解析輻射觀測資料來比對驗證結果。
研究結果顯示，在雲量演算法中使用光學氣膠厚度(Aerosol Optical Depth，AOD)來判定大氣灰濛程度，並設定不同的閾值計算，在高污染灰濛的影像下能得到比原廠雲量產品更精準的雲量。而本研究建立之輻射量模式與BSRN的輻射觀測數據有高度的一致性，其決定係數R2 有 0.94，且RMSE僅55.8W m-2，且觀測數據與模式結果的差值在100 W m-2以內的比例有達到93.1%，顯示此模式能有效掌握極短時輻射快速變化的情形。研究結果也發現透過不同情境之輻射量計算，能將短暫輻射高值的情境區分出來，後續能進一步針對此情境進行改進。而本研究建立之輻射量短期預測方式納入本模式參數的變化趨勢來進行預測，十分鐘內的預測決定係數R2都達0.81以上，RMSE大部分也都低於100 W m-2，其準確度在10分鐘內可達到80.3%，也說明本模式所建立之預測方式能夠掌握輻射短期的快速變化。上述結果在未來除了可應用在其他有架設全天空照相機之測站外，本模式也使用較低的運算資源即可進行，因此可搭配小型電腦與全天空相機模組至太陽能案場實際觀測與計算，提供太陽能相關業者極短時發電變化趨勢之參考。;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. |
