摘要
磁性元件在磁能传递、存储和滤波中起着关键作用,直接影响功率变换器的体积、质量、损耗及成本。因此,准确预测磁芯损耗至关重要。针对励磁波形对磁芯损耗的显著影响,提出了一种基于集成学习的励磁波形分类策略。首先,采用支持向量机(support vector machine,SVM)、随机森林(random forest,RF)和梯度提升决策树(gradient boosting decision tree,GBDT)作为基分类器,通过将分类结果与原始特征结合构建新的特征集,并使用元分类器进行训练以提升模型的泛化能力;然后,选择XGBoost作为磁芯损耗预测的核心模型;最后,通过遗传算法进行多目标优化,寻找到最小磁芯损耗与最大传输磁能的最佳工况。实验结果表明:提出的集成学习分类模型能够准确分类励磁波形,XGBoost模型相较于传统磁芯损耗预测模型及其他机器学习模型,展现了更高的预测精度和拟合效果。优化后的模型成功实现了磁芯损耗最小化与传输磁能最大化的平衡。
Magnetic components play a key role in energy transfer,storage,and filtering,directly affecting the size,weight,loss,and cost of power converters.Therefore,accurate prediction of core loss is essential.To address the significant influence of excitation waveforms on core loss,an ensemble learning-based waveform classification strategy is proposed.Support vector machine(SVM),random forest(RF),and gradient boosting decision tree(GBDT)are used as base classifiers.The classification outputs are combined with original features to construct a new feature set,which is then used to train a meta-classifier to enhance generalization.XGBoost is selected as the core model for core loss prediction.A genetic algorithm is applied for multi-objective optimization to identify the optimal operating condition with minimal core loss and maximal magnetic energy transfer.Experimental results show that the ensemble classification model can accurately classify excitation waveforms.Compared with traditional core loss prediction models and other machine learning methods,the XGBoost model demonstrates higher prediction accuracy and better regression performance.The optimized framework demonstrates the capability to meet both loss reduction and energy efficiency objectives.
作者
姚启达
平鹏
朱心怡
朱新凡
YAO Qida;PING Peng;ZHU Xinyi;ZHU Xinfan(School of Transportation and Civil Engineering,Nantong University,Nantong 226019,China)
出处
《南通大学学报(自然科学版)》
2025年第2期29-38,共10页
Journal of Nantong University(Natural Science Edition)
基金
国家自然科学基金青年科学基金项目(52202496)。
作者简介
第一作者:姚启达(2001-),男,硕士研究生;通信联系人:平鹏(1988-),男,副教授,博士,主要研究方向为人工智能、大数据分析。E-mail:pingpeng@ntu.edu.cn。
