甘景福, 马明晗, 李永刚.基于极限学习机的变压器热点温度实时预测方法研究[J].河南理工大学学报（自然科学版）,2021,40(2):135-141.

GAN J F, MA M H, LI Y G.Study on transformer hot spot temperature real-time prediction method based onextreme learning machine[J].Journal of Henan Polytechnic University(Natural Science) ,2021,40(2):135-141.

基于极限学习机的变压器热点温度实时预测方法研究

甘景福1, 马明晗2, 李永刚2

1.国网冀北电力有限公司 唐山供电公司，河北 唐山 063000;2.华北电力大学 电力工程系，河北保定 071000

摘要:油浸式变压器的负载能力和运行寿命与其热点温度密切相关，准确预测变压器热点温度对监测变压器健康状态和制定动态增容决策有着重要意义。为了预测变压器热点温度，以变压器负载率、环境温度和顶层油温为特征值，采用BP神经网络、Elman神经网络和极限学习机 （extreme learning machine ， ELM） 3种方法分别拟合变压器的热点温度，结果表明，ELM模型的拟合度更高，运算速度更快。通过温度、湿度、天气类型等7类变压器运行历史数据，构建基于支持向量回归（support vector regression ， SVR ）的电力负荷预测模型，将预测结果作为ELM模型的前置输入，提出一种基于极限学习机的变压器热点温度实时预测方法。以某220 kV油浸式变压器的运行数据为样本对该方法进行验证，发现预测值与真实值的误差在±1 ℃之内，表明该预测方法具有良好的预测精度。

关键词:油浸式变压器;热点温度预测;极限学习机;支持向量回归

doi:10.16186/j.cnki.1673-9787.2020010019

基金项目:国家自然科学基金资助项目（51777075 ）；国家电网公司科技项目（5207031801CR）

收稿日期:2020/01/07

修回日期:2020/05/15

出版日期:2021/03/15

Study on transformer hot spot temperature real-time prediction method based onextreme learning machine

GAN Jingfu1, MA Minghan2, LI Yonggang2

1.Tangshan Power Supply Company， State Grid Jibei Power Co. ， Ltd. ， Tangshan 063000 ， Hebei， China;2.Department of Electric Power Engineering ，North China Electric Power University ，Baoding  071000 ，Hebei， China

Abstract:The load capacity and operation life of oil-immersed transformer are closely related to its hot spot temperature. Accurate prediction of transformer hot spot temperature is of great significance for monitoring transformer health， it is also of great significance for making dynamic capacity increase decision. In order to predict the transformer hot spot temperature， the transformer load rate， the environment temperature and the top layer oil temperature were taken as the characteristic values to fit the transformer hot spot temperature. BP neural network， Elman neural network and extreme learning machine （ ELM ） were used respectively. The results showed that the ELM had higher fitting degree and faster operation speed in the three methods. Then， the power load prediction model based on support vector regression （ SVR） was constructed. Seven kinds of transformer operation history data， such as temperature， humidity and weather type， were used in this model. Taking the prediction results as the pre-input of ELM model， a real-time prediction method of transformer hot spot temperature based on ELM was proposed. The operation data of a 220 kV oil-immersed transformer was taken as a sample to verif the proposed method， and the prediction results of this method could basically ensure that the error between the predicted value and the real value was within ± 1 °C. This indicated that the prediction method had good prediction accuracy.

Key words:oil-immersed transformer;hot spot temperature prediction;extreme learning machine;support vector regression

 基于极限学习机的变压器热点温度实时预测方法研究_甘景福.pdf