Analysis of Influencing Factors of PV Based Ensemble Modeling for PV Power and Application in Prediction

摘要

According to the volatility and intermittent characteristics of photovoltaic power generation. Integrating PV power to the grid have an impact on the stability and safety. To address this challenge, the work learns the effect of support vector machine (SVM) and several algorithms on forecast. An algorithm model for improving the prediction accuracy of training data for multiple groups of factors has been proposed. The model consists of gradient boosting decision tree (GBDT), Particle Swarm Optimization (PSO) and SVM. Finally, according to the integrated algorithm, assigning weak learners' weights and integrating become strong learners. The GBDT algorithm is able to find the factors with high conelation coefficient in the data to establish the model, avoiding of using the empirical method to select the factors. The PSO algorithm finds the optimal parameters of the SVM algorithm and the optimal weight of the weak learner. Compared with BP and traditional SVM, the model established by the data without determining the weather type can obtain better prediction accuracy.