| dc.description.abstract | In recent years, due to the excessive use of the fossil energy, carbon emissions have caused the global climate warming. In order to reduce the carbon emissions, countries around the world committed to the development of green energy which includes solar power, wind power and hydropower. In Taiwan, research of the solar power gets more attention gradually. But the solar irradiance would change dramatically due to season, time, weather and occlusion of clouds. These factors may cause the worries on the reliability of the solar power system. And forecasting short-term irradiance is important for the operators to manage and allocate resources. In our research, we use the image processing technology to analyze the all-sky images, and use the analysis results to predict the occlusive situation between sun and clouds. The prediction would help increase the reliability of the short-term solar irradiance forecasting.
In our research, we read the all-sky images and detect the area of clouds in the images as a mask first. Then, we use the image difference to get the motion region. Applying the cloud mask to motion region, we can get the cloud motion region in the images. Afterwards, we use the cloud motion region to detect the feature points. Then the feature points will be clustered by a clustering algorithm. After obtaining the clustering results, we perform tracking of feature clusters in continuous images. After tracking, we use the tracking information to calculate the feature vector. Then, we use this vector to train the predictive model. Finally, we do the prediction and validate the results with the ground truth. And we get a good performance that the prediction accuracy is higher than 85%. | en_US |