基于PSO-SVM的光伏阵列故障检测与分类

光伏阵列故障的精确检测是提高光伏电站运行可靠性和安全性的重要因素之一。本文提出了粒子群优化支持向量机(Particle Swarm Optimization-Support Vector Machine, PSO-SVM)的光伏阵列故障检测与分类的方法。分析了光伏阵列输出特性和故障类型,选择合适的特征向量及归一化方式。选用径向基核函数优化模型结构,并利用PSO算法对参数进行寻优,提高模型精确度。结合实验平台,获取光伏阵列正常工作和8种故障状态的实测数据,随机划分为训练集和测试集,并建立PSO-SVM故障检测与分类模型。实验表明应用本文模型进行故障检测准确率达99.89%,分类准确率达98.68%,优于BP (Back Propagation)神经网络以及决策树的检测和分类结果。

Fault Detection and Classification in Photovoltaic Arrays Based on PSO-SVM

Accurate Fault detection in photovoltaic arrays is one of the key factors for increasing the reliability and safety of photovoltaic (PV) systems. The Particle Swarm Optimization-Support Vector Machine(PSO-SVM) modeling of fault detection and classification (FDC) in photovoltaic arrays is presented in this paper. The characteristic and fault patterns of the PV arrays are analyzed to select the appropriate feature vectors and normalization method. In order to strengthen the accuracy of the proposed model, the RBF kernel function is applied to improve the model structure, whose parameters are optimized by the PSO algorithm. Combined with the measured platform, the experiment data for the PV array under a normal working condition and eight types of faults are recorded. Data are randomly divided into testing set and training set to train the PSO-SVM model. The fault detection and the fault classification accuracies of the proposed model are 99.89% and 98.68% respectively, and are better than the accuracies of BP neural network and decision tree.