Matrix based univariate and multivariate linear similar day approach towards short term solar radiation forecasting

Increase in the penetration of renewable generation like solar and wind in power system requires expansion of transmission structure as well as more flexible generation, better cycling of existing generation, accurate spinning reserve planning, optimum power dispatch from renewable plants and security evaluations and stability analysis with high penetrated renewables. Almost all these operations require real time predicted generation from intermittent renewable sources. There are several models present in literature for solar power prediction, but accuracy and real time applications are the major challenges. Artificial neural network (ANN) models, Numerical Weather Prediction (NWP) models and Machine Learning models like Support Vector Regression, Bayesian Approaches and Hybrids models are used for prediction. The processing time of these models for decently accurate prediction is quite high. This paper investigate the similarity of solar radiation of similar days in different years and also implements one univariate and two multivariate similar day based linear models for short term solar power prediction. Similar days inputs are used prediction because, position of sun and earth is same corresponding to previous year similar day, only difference is cloud cover and cloud movement, wind speed and temperature. If we assume dependent factors will remain same with respect to previous years similar day, solar radiation will be same as previous years similar day values. Results obtained from these models shows that these models have strong potential for accurate real time solar power prediction and their accuracy can be enhanced by considering factors affecting solar irradiation.