Skip to Main content Skip to Navigation

Health monitoring of photovoltaic modules using electrical measurements

Abstract : Fault detection and diagnosis are essential elements for the condition monitoring of photovoltaic (PV) panels. This thesis proposes a new four-step strategy (modelling, pre-processing, extraction, and analysis of signatures) using full current-voltage characteristics (I-V curves). The modelling is based on an approach driven by simulated or measured data. For the pre-processing, to mitigate the effects of the different measurement conditions, we proposed an improved I-V curve correction procedure that is better adapted to defective panels than the standard ones. Besides, the current vector is resampled to have the same number of points. For feature extraction after this pre-processing, three methods are developed: direct use of the I-V feature or its transformation by the Gramian Angular Difference Field (GADF) or Recurrence Plot (RP) technique. Principal component analysis (PCA) is also applied to reduce the dimension of the feature matrix.For feature analysis, six common machine learning techniques are evaluated: artificial neural network (ANN), support vector machine (SVM), decision tree (DT), random forest (RF), k-nearest neighbour (kNN), and Bayesian naive classifier (NBC). To evaluate the different combinations of features and classifiers, the performance criteria used are classification accuracy and computational complexity. Eight conditions (one healthy and seven defective) of the PV panels are studied using simulated and measured I-V curves to build the database. The results show that using the features from the GADF transformation of the I-V curves as inputs to the ANN classifier achieves 100% classification accuracy for both simulated and measured data on a test bench developed in the laboratory. The robustness to perturbations, the impact of PCA and the feature transformation are also addressed. The proposed strategy is also compared to those that only partially use the I-V curve information and techniques in literature.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, November 10, 2021 - 6:25:10 PM
Last modification on : Friday, July 8, 2022 - 3:23:47 AM
Long-term archiving on: : Friday, February 11, 2022 - 7:54:14 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03425471, version 1


Baojie Li. Health monitoring of photovoltaic modules using electrical measurements. Electric power. Université Paris-Saclay, 2021. English. ⟨NNT : 2021UPAST087⟩. ⟨tel-03425471⟩



Record views


Files downloads