多层SVR选择机制在变流器噪声预测的应用

Application of Multi-layer SVR Selection Mechanism for Converter Noise Prediction

针对轨道交通车辆中的变流器噪声降噪问题,提出一种基于多层支持向量回归(Support Vector Regression,SVR)的选择机制,用以构建噪声预测模型。首先,以训练基准函数为支撑,构建特征向量与SVR、核函数之间的映射关系。随后根据测试数据的特征,依托之前的映射关系,完成SVR、核函数的筛选。最后,使用粒子群优化算法(Particle Swarm Optimization,PSO)完成参数匹配和模型构建。研究中先通过三个测试函数,验证多层SVR选择机制的准确率,再将构建的方法应用于变流器的噪声预测。结果表明:同其他常用的SVR方法相比,所研究的方法在预测效果上取得较大的提升。

A selection mechanism based on multi-layer Support Vector Regression(SVR)is proposed to construct a noise prediction model for the noise reduction problem of converters in rail vehicles.Firstly,a mapping relationship between the feature vector and SVR-kernel is constructed with the support of the training benchmark function.The SVR and kernel functions are then selected according to the characteristics of the test data based on the previous mapping relations.Finally,Particle Swarm Optimization(PSO)is used to complete the parameter matching and model construction.The accuracy of the multi-layer SVR selection mechanism is verified by three test functions,and then the method is applied to the noise prediction of the converter.The results show that this method achieves a significant improvement of prediction results in comparison with other commonly used SVR methods.