Study On Short-term Power Forecasting Of PV Power System

Photovoltaic power generation has been widely used becauce of these advances,such as less pollution,flexible scale,and so on.However,it is difficult for PV systems to implement energy management and grid scheduling effectively due to the uncertainty,the volatility and intermittent of the PV system affected by environmental factors.And the larger influce on PV systems by environmental factors,the more difficult to forecast the power output of PV systems accurately.So,it is important to forecast the power output of single-pointprediction and probability distribution prediction,which can both ensure power prediction accuracy and provide comprehensive PV ifnormation.It has important significance for security dispatching,stable operation and energy management.Based on the current development of photovoltaic power generation system,this paper carry out the sthdy on short-term power forecasting of PV power system.Firstly,to determin the main environmental factors that affect the PV short-term power,this paper analyze the relevace between the short-term power output and different weather factors,including weather patterns,light intensity and temperature.Then according to the determined weather factors,this paper confirms the similar day of the forecasting day.Taking the data of weather factors and power output of the similar day as the the input of the forecasting model,this paper establish the forecasting model based on the improved support vector machine algorithm(SVM)to forecast the single-point power and the up and down power of the given confidence level.This forecasting model has a high prediction accuracy.This paper also establishes the forecasting model based on restricted bolzmann machine improved by genetic algorithm(GA-RBM)to forcast the single-point power and the probability distribution.And based on the probability distribution and the given confidence level,this paper can get the confidence inteval.According to the deep learning theory,this paper establishes the forecasting model based on deep belief network(DBN)to forcast the single-point power and the probability distribution.This method has a higher prediction accuracy.And the simulation example has proved the reliability and accuracy of the forecasting model.