| Summary: | 碩士 === 國立中央大學 === 資訊工程學系 === 103 === Renewable energy is growing quickly in the modern society. Many countries have devoted themselves to the development of renewable power. And solar energy is one of the most important renewable energy. To overcome its unstable nature and achieve better utilization, forecasting short-term solar irradiance precisely is a crucial issue. This paper proposes a short-term irradiance prediction framework that based on automatic cloud classification. The cloud types are classified according to the features extracted from all-sky images. Multiple regression models are constructed by different cloud types using historical clearness indices or irradiance values as features. Moreover, ramp-down events are detected and the predicted irradiance is corrected on ramp-down events. The amount of correction is determined by the features extracted from the all-sky images. We also design a Kalman-filter based prediction model with time-varying system matrix. Afterwards, we fuse the prediction results of the regressor and the Kalman filter predictor. Finally, we validate the proposed system with two different datasets. Experiments have shown that incorporating cloud type information can capture different characteristics of irradiance variation under different cloud types. Also, the design of time-varying system matrix is able to improve the prediction accuracy.
|
|---|
