基于二次采样和集成学习方法的变压器故障预测
Transformer Fault Forecast Based on Re-sampling and Integrated Learning Approach
侯赛 1成润坤 1刘达1
作者信息
- 1. 华北电力大学经济与管理学院,北京 102206;华北电力大学智慧能源研究所,北京 102206
- 折叠
摘要
提前预测预警变压器故障可帮助电网企业提前安排检修,保证电网运行安全.提出了一种基于二次采样和集成学习的变压器故障预测模型,以增加故障数据样本并提高模型预测精度.首先,引入合成少数类过采样(SMOTE-SMOTE)二次采样方法处理变压器运行状态初始数据,构建平衡数据集,然后以支持向量机(SVM)、随机森林(RF)、人工神经网络(ANN)等为基学习器对平衡后的变压器运行状态数据集进行预测,构建Stacking集成学习模型对变压器故障进行预测.综合对比不同采样方法、不同基学习器和元学习器下模型的预测效果,结果表明二次采样方法相较于单一采样方法可以有效提高模型的预测效果.
Abstract
Early prediction and warning of transformer failures can help grid companies to schedule maintenance in advance to ensure safe grid operation.A transformer fault forecasting model based on re-sampling and integrated learning is proposed to increase fault data samples and improve model prediction accuracy.Firstly,the SMOTE-SMOTE quadratic sampling method is introduced to process initial transformer operating status data to construct the balanced dataset,and then SVM-RF-ANN and other base learners are used to predict the balanced ransformer operating status dataset and construct the Stacking integrated learning model to forecast transformer faults.The prediction effect of the models under different sampling methods,different base learners and meta learners are compared comprehensively,it is shown that the secondary sampling method can effectively improve the prediction effect of the model compared with single sampling method.
关键词
电力变压器/故障预测/二次采样/集成学习Key words
power transformers/fault forecasting/re-sampling/integrated learning引用本文复制引用
基金项目
国家重点研发计划资助项目(2020YFB1707800)
出版年
2024
NETL
NSTL
万方数据