區塊匹配應用於雲層移動估計之全天空影像與衛星雲圖日射量估計與預測系統

Cloud motion estimation based on block matching for solar irradiance estimation and prediction with all-sky images and satellite cloud images

高碩(Kao Shuo)

指導教授 ： 呂藝光

國立臺灣師範大學/科技與工程學院/電機工程學系/碩士(2024年)

https://doi.org/10.6345/NTNU202401627

摘要

本文透過全天空影像及衛星雲圖的分析，來開發以影像中雲層為特徵的估計與預測系統。在全天空影像中，使用紅藍比例法擷取雲層特徵影像，並提出適應式閾值在不同背景亮度的情況下更精確地判定雲層資訊計算雲層在整張圖的占比，以及太陽周圍雲特徵分析，在衛星雲圖中使用超像素分割出觀測區域雲層並利用紅藍比例法擷取雲層資訊；透過粒子測速區塊匹配推估雲層移動情形，製作未來數分鐘至小時之雲層情況，提取雲層特徵，作為雙向長短期記憶模型之輸入，而模型輸出為日射量。使用三個評估指標來檢視模型學習情形，包含相對均方根誤差、相對平均絕對誤差與預測技巧比較估計及預測成果，其中估計與預測時長 120 分鐘實驗的相對平均絕對誤差分別可達 21.99%與 34.66%。

關鍵字

衛星雲圖 ； 全天空影像 ； 粒子測速區塊匹配 ； 雙向長短期記憶

並列摘要

This thesis develops irradiance estimation and prediction system using all-sky and satellite cloud images. In all sky images, the red-blue ratio method and adaptive thresholds are used to extract cloud features and analyze cloud coverage. Superpixel segmentation is applied to extract cloud features in satellite cloud images. Particle image velocimetry block matching is used to compute cloud motion. Cloud features are fed into a bidirectional long short-term memory model to forecast solar irradiance. Performance is evaluated using relative root mean square error, relative mean absolute error, and forecast skill metrics. The relative mean absolute erros of the forcasting with lead time of 2 hours and estimation are 34.66% and 21.99%, respectively.

並列關鍵字

satellite cloud images ； all-sky images ； particle image velocimetry block matching ； bidirectional long short-term memory

參考文獻

[1] T. Hsiao, K. Huang, and Y. Chang, “A model for forecasting of solar power annual electricity and its strategy application,” Journal of Taiwan Energy., vol. 4, no. 4, pp. 401-430, 2017.

[2] M. Q. Raza, M. Nadarajah, and C. Ekanayake, “On recent advances in PV output power forecast,” Solar Energy, vol. 136, pp.125-144, Oct. 2016.

[3] J. Almorox, C. Voyant, N. Bailek, A. Kuriqi, and J. A. Arnaldo, “Total solar irradiance's effect on the performance of empirical models for estimating global solar radiation: An empirical-based review,” Energy, vol. 236, 2021.

[4] M. Z. Hassan, M. E. K. Ali, A. S. Ali, and J. Kumar, “Forecasting day-ahead solar radiation using machine learning approach,” in Asia-Pacific World Congress on Computer Science and Engineering. Mana Island, Fiji, Dec. 2017, pp. 252-258.

[5] X. Shao, S. Lu, and H. F. Hamann, “Solar radiation forecast with machine learning,” in International Workshop on Active-Matrix Flatpanel Displays and Devices. Kyoto, Japan, Aug. 2016, pp. 19-22.