A model using support vector regression (SVR) combined with particle swarm optimization (PSO) was employed to construct mathematical prediction model for relationship between magnetic properties and alloy compositions according to experimental sample data of NdFeB permanent magnets based on the uniform design method. The leave-one-out cross validation (LOOCV) test results show that dependence of magnetic properties on alloy compositions is very complicated and highly nonlinear. The mean absolute percentage error for Br, Hcj and (BH)max are 0.53%, 3.90%, 1.73%, and the correlation coefficient (R2) is as high as 0.839, 0.967 and 0.940, respectively. This investigation suggests that the PSO-SVR is an effective method to predict the properties of NdFeB magnet and resultantly provides theoretical guidance for researching dependence of magnetic properties on alloy composition for experimental researchers.%根据均匀设计方法制备的 NdFeB 系永磁合金样品实验数据,应用基于粒子群算法(PSO)寻优的支持向量回归(SVR)方法,建立了磁性能与合金成分之间预测模型。留一交叉法结果表明磁性能与合金成分之间关系复杂,呈现高度的非线性。表征磁性能的剩磁 Br、矫顽力 Hcj和磁能积(BH)max的平均绝对百分误差分别为0.53%、3.90%和1.73%,相关系数(R2)分别高达0.839、0.967和0.940。该方法有效地预测了NdFeB粘结磁体的磁性能,为实验工作者研究合金成分与磁性能之间关系提供了理论指导。
展开▼