PV Output Prediction Basedon Gradient Boosting Decision Tree Model With Bayesian Optimization Algorithm and Fine-grained Features

Main Article Content

Xie C., Wang J., Xie X., Liu Z., Bai J.

Abstract

Photovoltaic (PV) output is closely related to weather patterns.Deep mining of weather pattern features can effectively improve accuracy of PV output prediction. Since the degree of information granularity has influence on accuracy of PV output prediction, thetraditional PV output prediction methods using coarse-grained original features or clustering features have potential for improvement of prediction accuracy. For the above reasons, this paper proposes a PV output prediction method based on gradient boosting decision tree (GBDT) with Bayesian optimization algorithm (BOA) and fine-grained features.Firstly,the method constructs fine-grained features for each meteorological monitoring data and PV output monitoring datain the daytime, including instantaneous weather pattern features and time window trend features.Then the BOA is used to reduce the types of fine-grained features. Finally, the relationship between the features and PV output curves is fitted with the GBDT model and establishes a BOA-GBDT PV output prediction model. According to the error analysis of practical examples,the results show that compared with the traditional SVM(Support Vector Machine) method, the running time of theprediction model constructed by this methoddecreases by 97.3% and the root mean square error decreases by 80.4%. The effectiveness of this method is verified. 

Article Details

Section
Articles