A photovoltaic power prediction method for regions with frequent changes of meteorological characteristics

Abstract

Accurate prediction of photovoltaic （PV） power is of great significance to the stable operation of power grid. Therefore， a PV power prediction method is proposed based on frequent changes of meteorological features to improve the prediction accuracy. Firstly， the multivariate time series of PV power prediction is constructed based on Person correlation analysis. Secondly， multivariable phase space reconstruction （MPSR） is performed for the time series of PV power prediction by C-C method to further investigate the coupling between PV power and meteorological characteristics. Finally， support vector regression （SVR） is used for non-linear fitting and predicting the phase space after PV power and meteorological feature reconstruction. To verify MPSR can improve the prediction effect， the paper compares MPSR that combines with back propagation neural network （BPNN） and radial basis function neural network （RBFNN）. Example analysis shows that MPSR can further explore the feature information contained in the regions with frequent changes of meteorological features. The prediction model that combines with MPSR improves the prediction accuracy of PV power.

Keywords

• photovoltaic power prediction
• characteristics mining
• multivariable phase space reconstruction
• support vector regression