摘要
建立一种粒子群优化的改进加权支持向量回归的变压器顶层油温模型,能够准确的估计变压器顶层油温。该模型根据环境温度、变压器负荷、变压器顶层油温等样本数据建立支持向量回归模型,分析变压器顶层油温与其他各因素之间的联系,根据不同影响因素建立支持向量超平面将变压器顶层油温预测限制在一个合理区间,根据支持向量机的惩罚因子和松弛因子的选择使该区间缩小至与实际变压器顶层油温的误差达到最小,使以变压器顶层油温为预测目标函数的预测模型精度最高。在支持向量回归模型建立时采用粒子群算法对其惩罚因子和松弛因子进行寻优以使支持向量回归模型预测效果达到最优。通过主成分分析方法对核函数进行改进从而优化支持向量回归模型,相比粒子群优化的支持向量机考虑数据特征量的权重,预测结果准确率更高。该模型利用支持向量回归方法不需要大量样本、不涉及概率测度、能够处理多维影响因素等优点,能够应对变压器油温短期预测数据不足或采集的油温相关数据维度较多的情况,给出准确的顶层油温预测结果。
A kind of top oil temperature model of transformer of the improved weighted support vector regression(SVR)based on particle swarm optimization(PSO)is set up,which can accurately estimate the top oil temperature of transformer. This model sets up SVR model in accordance with such sample data as ambient temperature,transformer load and top oil temperature of transformer,analyzes the relationship between the top oil temperature of transformer and other factors,sets up the support vector hyperplane in accordance with different influencing factors and limits the prediction of the top oil temperature of transformer to a reasonable interval. According to the choice of penalty factor and relaxation factor of the support vector machine,the error between this interval and the actual oil temperature of the top layer of transformer is minimized so to keep the predicted model accuracy,for which the top oil temperature of the transformer is as the prediction function,is the highest. During the setup of SVR model,the particle swarm algorithm is used to optimize the penalty factor and relaxation factor so to keep the prediction result of the SVR model reach optimal. The kernel function is improved by principal component analysis so to optimize the support vector regression model. Compared with particle swarm optimization support vector machine(SVM)considering the weight of data feature quantity,the accuracy of the prediction result is higher. This model uses such advantages of support vector regression method as not requiring a large number of samples,not involving probability measure and being able to deal with multi-dimensional influencing factors etc,can provide accurate top-oil temperature prediction results in case of insufficient short-term prediction data of transformer oil temperature or more dimensions of oil temperature related data collected.
作者
李诗勇
薛静
吴冕之
谢荣斌
靳斌
张鸿儒
李清泉
LI Shiyong;XUE Jing;WU Mianzhi;XIE Rongbin;JIN Bin;ZHANG Hongru;LI Qingquan(Guizhou Power Grid Co.,Ltd.,Guiyang Power Supply Bureau,Guiyang 550081,China;Shandong Provincial Key Laboratory of UHV Transmission Technology and Equipment,School of Electrical Engineering,Shandong University,Jinan 250061,China)
出处
《高压电器》
CAS
CSCD
北大核心
2021年第12期103-109,共7页
High Voltage Apparatus
基金
国家自然科学基金资助项目(51777117)
贵州电网有限责任公司科技资助项目(060100KK52170127)。
关键词
电力变压器
顶层油温预测
粒子群优化
加权支持向量机
power transformer
top-oil temperature prediction
particle swarm optimization
weighted support vector
