Fault diagnosis approach for photovoltaic array based on the stacked auto-encoder and clustering with I-V curves

Photovoltaic arrays are usually installed outdoors in harsh environments and prone to various faults, which will seriously affect the efficiency of photovoltaic arrays. Therefore, the effective fault detection and diagnosis plays an important role in the safe, operation, and maintenance of the photovoltaic plant. In recent years, machine learning methods have made remarkable achievements in fault diagnosis. However, there still exist some limitations: (1) feature extraction relies on expert experience and lacks automation. (2) artificial feature extraction easily ignores some potential useful features. (3) the nonlinear characteristics of current-voltage curves cannot be effectively learned by the shallow network structure. In order to address the above issues, the supervised deep learning methods with automatic feature extraction capability are applied, but a lot of labeled data are required for pre-training. Therefore,a fault diagnosis method is proposed for photovoltaic array based on stacked autoencoder and clustering algorithm in this paper, which can automatically extract features and use a small number of labeled data samples to mine data sample features for fault diagnosis. Firstly, the effective features are automatically extracted by the stacked auto-encoder from the current-voltage curves. Secondly, the dimension of the features is reduced and visualized by the t-distributed stochastic neighbor embedding to improve the performance of the clustering algorithm. Finally, clustering centers and clusters are obtained by clustering algorithm and membership function is used for fault diagnosis. Moreover, the simulation and experimental data are used to verify the performance of the proposed fault diagnosis method. The 97.3% and 98.3% classification accuracies are obtained in the simulation and experimental results.

Automated summary

A fault diagnosis method for photovoltaic arrays based on deep learning methods is proposed, which can automatically extract features and use a small number of labeled data for fault diagnosis.