基于IWOA-SVM的变压器绕组热点温度预测
Transformer winding hot spot temperature prediction based on IWOA-SVM
马成 1罗亭然 1刘闯 2卢银均2
作者信息
- 1. 国网湖北省电力有限公司荆州供电公司,湖北 荆州 434000
- 2. 国网湖北省电力有限公司荆门供电公司,湖北 荆门 448000
- 折叠
摘要
为了降低变压器高温运行风险和提高绕组热点温度预测精度,提出了一种基于改进鲸鱼算法优化支持向量机的绕组热点温度预测方法.采用灰色关联分析结果确定负载电流、有功功率、顶层油温和环境温度为引起绕组热点温度变化的主要特征量,并以此作为绕组热点温度预测模型的支持向量.利用余弦调整控制因子和引入自适应权重系数 2 种策略对鲸鱼算法进行改进,提高了改进鲸鱼优化算法(improved whale optimization algorithm,IWOA)的优化性能,采用IWOA算法优化支持向量机(support vector machine,SVM)参数,建立了基于IWOA-SVM的变压器绕组温度预测模型.算例分析结果表明,本文绕组热点温度预测方法的均方根误差为1.21℃、决定系数为0.897,平均相对误差为 2.14%,三项指标均优于其他方法,验证了所提方法的实用性和有效性.
Abstract
To reduce the risk associated with the high-temperature operation of transformers and to improve the accuracy of winding hot spot temperature prediction,this study proposes a method based on the improved whale optimization algorithm(IWOA)combined with an optimized support vector machine(SVM).As determined through grey correlation analysis,key influencing factors such as load current,active power,top oil temperature,and ambient temperature are identified as the main characteristic variables affecting winding hot spot temperature changes.These factors are utilized as support vectors for the winding hot spot temperature prediction model.The whale algorithm is refined by incorporating a cosine adjustment for control factors and introducing adaptive weight coefficients,which enhance the optimization performance of the IWOA.The SVM parameters are optimized using the IWOA algorithm,establishing an IWOA-SVM-based transformer winding temperature prediction model.Results from case studies show that the proposed method's root mean square error is 1.21℃,determination coefficient is 0.897,and average relative error is 2.14%.All three indica-tors surpass other methods in performance.This validation underscores the practicality and effectiveness of the proposed method.
关键词
变压器/绕组热点温度/改进鲸鱼算法/支持向量机/灰色关联分析Key words
transformer/winding hot spot temperature/improved whale optimization algorithm/support vector machine/grey correlation analysis引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据