Short-Term photovoltaic system power forecasting based on ECSVM optimized by GA
Yun-jun Yu, Yun-tao Xue, Sui Peng, Chao Tong, Zi-heng Xu
College of Information Engineering, Nanchang University Xuefu Rd. 999, Nanchang, Jiangxi, 330031, China
It is of great significance to research PV forecasting techniques for mitigating the effects of the randomness of the Photovoltaic output. This paper analyses many factors from PV which impact photovoltaic output and extracts the main factors, forming sample data combined with the historical database generation data from PV monitoring system. And an error correction SVM method (ECSVM) is used to calculate the open integration of photovoltaic power storage system in advance or after the time in order to try to eliminate the system error between the predicted and actual values. At the same time, using genetic algorithm to optimize kernel function parameter and the error penalty factor and other parameters in this model, the establishment of the GA-ECSVM model improves portfolio optimization model parameter prediction accuracy and efficiency of the selected type. Finally examples verified and compared with standard SVM methods and ECSVM method, predicting effects show that: The GA-ECSVM optimization model presented in this paper has better learning ability and generalization ability in the short term prediction of photovoltaic power generation, with the prediction accuracy of 95.2016%.