Short-term solar radiation prediction based on SVM with similar data

Solar energy as a clean and renewable energy gets more and more attention by the international community. Photovoltaic (PV) power generation connected grid is the main development tendency for utilizing solar energy in recent years. Due to discontinuity of PV conversion technology, stability of grid -connected PV power generation is challenged and PV power prediction becomes an effective way to solve these problems. Because the prediction accuracy for solar radiation is the first key problem for the PV power prediction, a solar radiation prediction method based on support vector machine (SVM) with similar data was proposed in the paper. Similar data was extracted from historical data by using pattern recognition with Euclidean distance to create the training samples. Employing to wavelet decomposition, the original solar radiation signal was decomposed into trend signal of low frequency band and random signal of high frequency band. Different SVM radiation prediction models were trained respectively and combined to obtain the final forecasting results. The simulation results show that similar data enhance the relevance of the data and improve the model prediction accuracy Wavelet decomposition reduces the non-stationary parts of solar radiation signals. Different SVM models better approximate the solar radiation characteristics of low and high frequency band and good prediction accuracy is obtained.