Study Of The Related Factors Of Distributed PV Output And Short Term Forecasting

The output of the PV generation is random and intermittent.With the large-scale integration of distributed photovoltaic systems into the grid,it is bound to make a huge challenge to the power grid dispatching management and security.PV output prediction is one of the key technologies to solve this problem.And that is the foundation of distributed photovoltaic power management,safety and scheduling.It also has important significance for improving the quality of photovoltaic grid.In order to study the technology of PV output forecasting,a remote on-line monitoring system for meteorological parameters and electrical parameters was designed.On the basis of this,the influence factors of PV output and the correlation of input parameters were studied.The existing prediction model had the shortcomings of high input dimension and complex structure,and the predicted value of the models were difficult to meet the practical requirement.In this paper,the theoretical data was used as an exogenous input variable,and the high frequency data was obtained by the wavelet analysis as the characteristic data,so as to simplify the classification.In addition,to reduce the coupling degree of the data by calculation of related factors.Then Gamma Test and multi-objective genetic algorithm combination algorithm was used to achieve intelligent selection of input variables and reduce the redundancy of information.Then a support vector regression machine model,a neural network model and a auto regressive model were created to improve the accuracy of prediction,and to achieve the average output power classification.The relative RMSE were 9.7%,9.1%,7.8% in the low weather changing condition,and the relative RMSE were 13.54%,13.36%,13.87% in another case.The simulation results showed that the improvement measures and the correction methods had a significant effect on improving the prediction accuracy and reducing the training time.Compared to the pre optimization accuracy,that was more than 3%.At the same time,there was still a lot of uncertainty in the prediction of photovoltaic power generation,and single point prediction was difficult to meet all requirements.In this paper,the 2D interval prediction model was established,which could be used to control the balance of load in power network and the dynamic management of energy storage.The average relative error of the basic algorithm model was 13.2% and the average error of support vector machine model was 11.04%.In this paper,a remote on-line monitoring system was developed for the requirement of fine management of distributed PV station.With the theoretical calculation data and the comprehensive characteristic variables,the information was made the best of.Then the Gamma Test and multi-objective genetic algorithm was used to improve the prediction accuracy of the model.In addition,as these forecast models were pool in bad weather condition,the 2D interval prediction model was proposed.This research can be used to achieve the coordination and cooperation between the PV system and the grid,as well as the economic optimization of photovoltaic system.