摘要
支持向量机(SVM)能较好地解决小样本、非线性特征的多分类问题,适用于电力变压器运行状态评估,但参数选择对分类效果有着显著影响。利用改进的粒子群算法(PSO)对支持向量机(SVM)参数进行寻优,通过引入收敛因子、惯性因子动态化和自适应粒子变异三种方法对传统的PSO算法进行改进,从而获得最佳的分类模型。该模型以变压器油中溶解气体浓度为评估指标,将变压器分成优秀、一般、良好、注意、故障等5个等级,其中故障又分为低温过热、中温过热、高温过热、低能放电、高能放电、局部放电6个类别。通过实例数据分析得出,用改进后的PSO算法优化得到的SVM分类器能对变压器的各种状态进行较准确的评估。
The Support Vector Machine (SVM) have a better solution in small sample problem and nonlinear characteristics of the multi-classification. As a result SVM is suitable for condition assessment of power transformer operation ; however the parameters of Support Vector Machine (SVM) have significant implications on the classification results. In order to obtain the best classification model, an improved particle swarm optimization (PSO) algorithm is introduced to optimize the parameters of the support vector machine (SVM). The model is based on transformer dissolved gas analysis (DGA) technique as evaluation method, the running states of transformer are divided into excellent, good, normal, attention and fault five levels, where the fault level is divided into low-temperature failure of overheating, medium-temperature failure of overheating, high-temperature failure of overheating , low energy discharge, high energy discharge and partial discharge six categories. By the analysis of sample data, we prove that using the improved PSO algorithm to optimize the SVM classifier can increase the state assessment accuracy of transformer.
出处
《电力科学与工程》
2011年第3期27-31,共5页
Electric Power Science and Engineering