Fault diagnosis approach for photovoltaic arrays based on unsupervised sample clustering and probabilistic neural network model

Abstract

Photovoltaic power plants which work under actual conditions are composed of a large quantity of photovoltaic arrays. The complex configuration of such photovoltaic arrays frequently produces various types of fault which directly affect the safe and economic operation of the power plant. There are still some problems for fault diagnosis of photovoltaic arrays: it is difficult to precisely represent the electrical characteristics of an array under different fault conditions, fault diagnosis models require the accurate division of fault samples, but the classification of fault data depends on artificial prior knowledge. This paper proposes a fault diagnosis method wherein: (i) the photovoltaic array output characteristics and distribution of electrical eigenvectors under typical fault conditions are be effectively analyzed; (ii) the per unit method and the Gaussian kernel function are introduced into the fuzzy C means algorithm to improve the applicability and fuzzy clustering ability of the unsupervised screen for various fault samples; and (iii) a probabilistic neural network fault diagnosis model is built with clustered data as the input. Practical operation data is used to successfully validate the effectiveness and feasibility of the proposed method.