Rapid and accurate modeling of PV modules based on extreme learning machine and large datasets of I-V curves. (15th June 2021)

• Record Type: Journal Article
• Authors: Chen, Zhicong; Yu, Hui; Luo, Linlu; Wu, Lijun; Zheng, Qiao; Wu, Zhenhui; Cheng, Shuying; Lin, Peijie
• Is Part Of: Applied energy. Volume 292(2021)
• Publication Date: 2021-06-15
• Subjects: PV modules -- PV modeling -- I-V characteristics -- Extreme learning machine -- Machine learning; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2021.116929
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: An extreme learning machine based rapid modeling method is proposed for PV modules. The input features and single-hidden layer feedforward neural network are optimized. Anomaly detection and grid resampling methods are proposed to preprocess datasets. The ELM based PV model is compared with BPNN, GRNN, SVM, RF and RNN based PV models. The ELM based PV model is greatly superior in terms of training speed. Abstract: Efficient and accurate photovoltaic (PV) modeling plays an important role in optimal evaluation and operation of PV power systems. Using current–voltage (I-V) curves measured at different operating conditions, a novel extreme learning machine (ELM) based modeling method is proposed for characterizing the electrical behavior of PV modules, which features high training speed and generalization performance. Firstly, a voltage-current grid based method is used to resample each raw measured I-V curve for reducing data redundancy of I-V curves, a slope change based detection method is proposed to exclude the abnormal I-V curves for improving the data quality, and an irradiance-temperature grid based method is applied to downsample the dataset. Secondly, a single hidden-layer feedforward neural network (SLFN) is proposed as the model structure, which is then trained by the ELM learning algorithm. Particularly, the configuration of the ELM is optimized by cross-validation. Finally, the proposed ELM based PV modeling method is verified and tested on the preprocessed large datasets of I-V curves of six PV modules from the National Renewable Energy Laboratory. Moreover, in order to verify the advantage, the ELM based PV modeling is further compared with some commonly used machine learning based methods. Experimental results demonstrate that the proposed ELM based PV modeling method features fast training, high accuracy and generalization performance. The comparison results further indicate that the ELM based model is greatly superior in terms of the training time, and is slightly better than other algorithms in terms of the model accuracy and generalization performance.