基于特征迁移的永磁同步电机性能预测

Performance prediction of permanent magnet synchronous motors based on feature transfer

为解决电工装备在日趋复杂应用环境下的性能分析与优化问题,通过挖掘设计、生产和性能历史数据,学习得到并应用隐含在产品和工艺技术中的数据特征和知识经验,是基于数据驱动的电工装备性能分析与优化的重要研究内容。因此提出一种智能自学习新方法,利用历史电机数据中学习知识和特征,迁移应用到新电机的性能分析中,通过领域自适应算法中的特征迁移,提取源域和目标域的特征到公共空间,然后对提取的特征进行对齐,使用历史电机样本数据建立的预测模型,用于新电机的性能预测中。实验表明,在不同的数据集上,电机齿槽转矩的预测精确度分别提高了64%和80%,电机效率的预测精确度分别提高了69%和82%,为电工装备的智能设计与优化提供了新的思路和方法。

In order to solve the problem of performance analysis and optimization of electrical equipment in the increasingly complex application environment, by mining historical design, production and performance data, the data characteristics and knowledge experience implicit were learned and applied in products and process technology which is based on data-driven, that is the important research content of performance analysis and optimization of electrical equipment. Therefore, a new intelligent self-learning method was proposed to try to learn knowledge and features from historical motor data, and transfer them to the performance analysis of new motors. The main method is to extract the source and target domains through feature transfer in the domain adaptive algorithm. The features were transferred to the public space, and then the extracted features were aligned, and the prediction model established using historical motor sample data was used in the performance prediction of the new motor. Experiments show that on different data sets, the prediction accuracy of motor cogging torque is increased by 64% and 80%, respectively, and the prediction accuracy of motor efficiency is increased by 69% and 82%, respectively, which provides new ideas and methods for the intelligent design and optimization of electrical equipment.