Temperature & Humidity-Weighted-Modified GRNN Based Prediction Model of Photovoltaic Power Generation

Weather factors directly affect photovoltaic (abbr PV) power generation, in order to improve the prediction accuracy of PV power generation, this paper proposed a modified prediction GRNN for which with temperature and humidity weighted. The prediction model was built according to different type of weather separately, first order function and exponential function for temperature and humidity are weighted to modify the input neurons of GRNN model, and grey correlation degree function (abbr GCDF) is used to select single-day and multi-day training data sets for point prediction and hour prediction respectively. To improve the stability of the prediction model, a coefficient optimization method based on relative standard deviation is proposed. The experimental results showed that: (1) The prediction accuracy of the modified GRNN model weighted by temperature and humidity is significantly higher than that of the unmodified GRNN model, and all of the root mean square error of point prediction corresponding to the optimal GRNN model for four kinds of weather is less than 1.2238%; (2) The GRNN with exponential function for temperature and humidity has the best fitness for point prediction, the GRNN with first order function for temperature and humidity fit best instead for the hour prediction.