Adouni Amel, Chariag Dhia, Diallo Demba, Ben Hamed Mouna, Sbita Lassaâd
Laboratoire Systèmes Photovoltaïques, Eoliens et Géothermaux, Ecole National d'Ingénieur de Gabes, Univérsité de Gabes, Tunisia.
Group of Electrical Engineering - Paris (GEEPS), (CNRS, Centrale Supélec, UPMC, Univ. Paris-Sud), 91192 Gif Sur Yvette, France.
ISA Trans. 2016 Sep;64:353-364. doi: 10.1016/j.isatra.2016.05.009. Epub 2016 Jun 2.
As per modern electrical grid rules, Wind Turbine needs to operate continually even in presence severe grid faults as Low Voltage Ride Through (LVRT). Hence, a new LVRT Fault Detection and Identification (FDI) procedure has been developed to take the appropriate decision in order to develop the convenient control strategy. To obtain much better decision and enhanced FDI during grid fault, the proposed procedure is based on voltage indicators analysis using a new Artificial Neural Network architecture (ANN). In fact, two features are extracted (the amplitude and the angle phase). It is divided into two steps. The first is fault indicators generation and the second is indicators analysis for fault diagnosis. The first step is composed of six ANNs which are dedicated to describe the three phases of the grid (three amplitudes and three angle phases). Regarding to the second step, it is composed of a single ANN which analysis the indicators and generates a decision signal that describes the function mode (healthy or faulty). On other hand, the decision signal identifies the fault type. It allows distinguishing between the four faulty types. The diagnosis procedure is tested in simulation and experimental prototype. The obtained results confirm and approve its efficiency, rapidity, robustness and immunity to the noise and unknown inputs.
根据现代电网规则，风力涡轮机即使在出现诸如低电压穿越（LVRT）等严重电网故障时也需要持续运行。因此，已开发出一种新的低电压穿越故障检测与识别（FDI）程序，以便做出适当决策，从而制定便捷的控制策略。为了在电网故障期间获得更好的决策并增强故障检测与识别能力，所提出的程序基于使用新型人工神经网络架构（ANN）的电压指标分析。实际上，提取了两个特征（幅度和相位角）。它分为两个步骤。第一步是故障指标生成，第二步是用于故障诊断的指标分析。第一步由六个人工神经网络组成，用于描述电网的三个相（三个幅度和三个相位角）。关于第二步，它由一个单一的人工神经网络组成，该网络分析指标并生成描述功能模式（正常或故障）的决策信号。另一方面，决策信号识别故障类型。它允许区分四种故障类型。该诊断程序在仿真和实验原型中进行了测试。获得的结果证实并认可了其效率、快速性、鲁棒性以及对噪声和未知输入的免疫力。
