利用全天空影像之雲追蹤預測短期太陽輻射;Cloud Tracking Using All-Sky Images for Short-Term Solar Irradiance Prediction

NCUIR > College of Earth Sciences > Graduate Institute of Space Science > Department of Earth Sciences > Item 987654321/97419

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/97419

Title:	利用全天空影像之雲追蹤預測短期太陽輻射;Cloud Tracking Using All-Sky Images for Short-Term Solar Irradiance Prediction
Authors:	陳羿彣;Chen, Yi-Wen
Contributors:	太空科學與工程學系
Keywords:	太陽輻射預測;全天空影像;雲追蹤;雲配對;全天空輻射;Solar Irradiance Prediction;All Sky Image;Cloud Tracking;Cloud Pairing;Global Horizontal Irradiance
Date:	2025-08-14
Issue Date:	2025-10-17 11:16:42 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨太陽能發電佔比逐年增加，如何準確預測太陽能發電量已成為關鍵議題，而太陽輻射量的短期預測是其中的核心挑戰。現有方法多高度仰賴運算能力或大量訓練資料，限制其在即時運用之可行性。本研究針對雲遮蔽太陽所引起的輻射量劇烈變化，提出一套低運算成本的即時預測方法。此輕量化的雲配對及雲預測演算法，採用外延法推估未來 1∼10 分鐘後雲厚度分布，進而預測輻射值。因架構簡潔、演算法複雜度低，可在消費級中階運算設備上運行，平均可於 2.23 秒內完成一次預測。透過實測資料評估，本方法能有效預測未來 10 分鐘內之輻射量趨勢，預測與實際影像輻射值呈現高度正相關 ( 1 分鐘預測： R > 0.91， nRMSE 為 10.04%； 10 分鐘： R=0.84， nRMSE 為 13.39%）。整體表現勝過既有模型，本方法在不依賴複雜訓練模型與高效能硬體的情況下，仍能提供高準確度的短期太陽輻射預測，提供一實用的短時間預測方案。;The share of solar power generation in the energy mix is steadily increasing. Accurate forecasting of solar output has therefore become essential. A key challenge in this task is short-term prediction of solar irradiance. Existing approaches often depend on high computational capacity or large training datasets, which limits their applicability in real-time operations. Rapid irradiance fluctuations caused by cloud obstruction is a significant challenge for solar energy modulation. This paper proposes a novel, lowcost, real-time prediction method to address this issue. The proposed method utilizes a lightweight cloud-matching and irradiance-predicting algorithm. An extrapolation approach is employed to estimate cloud thickness distribution 1– 10 minutes ahead, which is then used to predict irradiance values. Due to its simple architecture and low algorithmic complexity,the method can be implemented on consumer-grade mid-level computing devices, with each forecast taking an average of 2.23 seconds. Experiments with measured data show that the method accurately forecasts irradiance trends. For a 1-minute forecast, the method achieves a high correlation (R>0.91) with a normalized root mean square error (nRMSE) of 10.04%. A 10-minute forecast yields R=0.84 and nRMSE=13.39%. Compared with existing models, our approach avoids complex training procedures and reliance on high-performance hardware. This work presents a practical and accurate solution for short-term solar irradiance predicting, demonstrating a clear advantage in implementation complexity and computational cost.
Appears in Collections:	[Graduate Institute of Space Science] Department of Earth Sciences

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	3	View/Open

社群 sharing

Loading...